Download as
Browse Code »

Commit 0e854e1e1dab832696362b8210d202475bd32de7

Authored by Erickson Silva
1 parent e1cf2b2f
Exists in master and in 1 other branch devel

Remove arquivos binários (dependências) desnecessarios

Showing 22 changed files with 0 additions and 9606 deletions   Show diff stats
aelius-install.tar.gz
View file @e1cf2b2
No preview for this file type
aelius_install.sh
View file @e1cf2b2
@@ -1,36 +0,0 @@ @@ -1,36 +0,0 @@
1 -#!/bin/bash  
2 -  
3 -#Instalador do Tradutor Aelius  
4 -  
5 -cd ~/vlibras-core  
6 -  
7 -echo "\n# Extraindo...\n"  
8 -tar -xf aelius-install.tar.gz -C ~/.  
9 -  
10 -echo "# Instalando dependências...\n"  
11 -sudo apt-get install python-dev python-yaml python-numpy python-matplotlib  
12 -  
13 -cd ~/nltk-2.0.1rc1/  
14 -  
15 -sudo python setup.py install  
16 -  
17 -echo "\n# Finalizando...\n"  
18 -  
19 -# path to HunPos binaries  
20 -echo "\nPATH=\"${PATH}:$HOME/Applications/bin\"" >> ~/.bashrc  
21 -echo "export PATH\n" >> ~/.bashrc  
22 -  
23 -# path to Aelius and Translate package  
24 -echo "PYTHONPATH=\"${PYTHONPATH}:$HOME/Applications:$HOME/vlibras-core/tradutor/src/py\"" >> ~/.bashrc  
25 -echo "export PYTHONPATH\n" >> ~/.bashrc  
26 -  
27 -cd ..  
28 -  
29 -sudo mv usr-local-bin.tar.gz /usr/local/bin/  
30 -  
31 -cd /usr/local/bin/  
32 -  
33 -sudo tar -xf usr-local-bin.tar.gz  
34 -sudo rm usr-local-bin.tar.gz  
35 -  
36 -echo "### Instalação finalizada! \n## Execute o seguinte comando para concluir:\n\n$ source ~/.bashrc\n"  
recognize/src/julius/doc/Juliusbook-4.1.5.pdf
View file @e1cf2b2
No preview for this file type
recognize/src/julius/doc/manuals/accept_check.txt
View file @e1cf2b2
@@ -1,80 +0,0 @@ @@ -1,80 +0,0 @@
1 - accept_check  
2 -  
3 -ACCEPT_CHECK(1) ACCEPT_CHECK(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - accept_check  
9 - - Check whether a grammar accept / reject given word sequences  
10 -  
11 -SYNOPSIS  
12 - accept_check [-t] [-s spname] [-v] {prefix}  
13 -  
14 -DESCRIPTION  
15 - accept_check is a tool to check whether a sentence can be accepted or  
16 - rejected on a grammar (prefix.dfa and prefix.dict). The sentence should  
17 - be given from standard input. You can do a batch check by preparing all  
18 - test sentence at each line of a text file, and give it as standard  
19 - input of accept_check.  
20 -  
21 - This tool needs .dfa, .dict and .term files. You should convert a  
22 - written grammar file to generate them by mkdfa.pl.  
23 -  
24 - A sentence should be given as space-separated word sequence. It may be  
25 - required to add head / tail silence word like sil, depending on your  
26 - grammar. And should not contain a short-pause word.  
27 -  
28 - When a word belongs to various category in a grammar, accept_check will  
29 - check all the possible sentence patterns, and accept it if any of those  
30 - is acceptable.  
31 -  
32 -OPTIONS  
33 - -t  
34 - Use category name as input instead of word.  
35 -  
36 - -s spname  
37 - Short-pause word name to be skipped. (default: "sp")  
38 -  
39 - -v  
40 - Debug output.  
41 -  
42 -EXAMPLES  
43 - An output for "date" grammar:  
44 -  
45 - % echo '<s> NEXT SUNDAY </s>' | accept_check date  
46 - Reading in dictionary...  
47 - 143 words...done  
48 - Reading in DFA grammar...done  
49 - Mapping dict item <-> DFA terminal (category)...done  
50 - Reading in term file (optional)...done  
51 - 27 categories, 143 words  
52 - DFA has 35 nodes and 71 arcs  
53 - -----  
54 - wseq: <s> NEXT SUNDAY </s>  
55 - cate: NS_B (NEXT|NEXT) (DAYOFWEEK|DAYOFWEEK|DAY|DAY) NS_E  
56 - accepted  
57 -  
58 -  
59 -SEE ALSO  
60 - mkdfa.pl ( 1 ) ,  
61 - generate ( 1 ) ,  
62 - nextword ( 1 )  
63 -  
64 -COPYRIGHT  
65 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
66 -  
67 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
68 -  
69 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
70 - Technology  
71 -  
72 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
73 - Technology  
74 -  
75 -LICENSE  
76 - The same as Julius.  
77 -  
78 -  
79 -  
80 - 10/02/2008 ACCEPT_CHECK(1)  
recognize/src/julius/doc/manuals/adinrec.txt
View file @e1cf2b2
@@ -1,130 +0,0 @@ @@ -1,130 +0,0 @@
1 - adinrec  
2 -  
3 -ADINREC(1) ADINREC(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - adinrec  
9 - - record audio device and save one utterance to a file  
10 -  
11 -SYNOPSIS  
12 - adinrec [options...] {filename}  
13 -  
14 -DESCRIPTION  
15 - adinrec opens an audio stream, detects an utterance input and store it  
16 - to a specified file. The utterance detection is done by level and  
17 - zero-cross thresholds. Default input device is microphone, but other  
18 - audio input source, including Julius A/D-in plugin, can be used by  
19 - using "-input" option.  
20 -  
21 - The audio format is 16 bit, 1 channel, in Microsoft WAV format. If the  
22 - given filename already exists, it will be overridden.  
23 -  
24 - If filename is "-" , the captured data will be streamed into standard  
25 - out, with no header (raw format).  
26 -  
27 -OPTIONS  
28 - adinrec uses JuliusLib and adopts Julius options. Below is a list of  
29 - valid options.  
30 -  
31 - adinrec specific options  
32 - -freq Hz  
33 - Set sampling rate in Hz. (default: 16,000)  
34 -  
35 - -raw  
36 - Output in raw file format.  
37 -  
38 - JuliusLib options  
39 - -input {mic|rawfile|adinnet|stdin|netaudio|esd|alsa|oss}  
40 - Choose speech input source. Specify 'file' or 'rawfile' for waveform  
41 - file. On file input, users will be prompted to enter the file name  
42 - from stdin.  
43 -  
44 - 'mic' is to get audio input from a default live microphone device,  
45 - and 'adinnet' means receiving waveform data via tcpip network from  
46 - an adinnet client. 'netaudio' is from DatLink/NetAudio input, and  
47 - 'stdin' means data input from standard input.  
48 -  
49 - At Linux, you can choose API at run time by specifying alsa, oss and  
50 - esd.  
51 -  
52 - -lv thres  
53 - Level threshold for speech input detection. Values should be in  
54 - range from 0 to 32767. (default: 2000)  
55 -  
56 - -zc thres  
57 - Zero crossing threshold per second. Only input that goes over the  
58 - level threshold (-lv) will be counted. (default: 60)  
59 -  
60 - -headmargin msec  
61 - Silence margin at the start of speech segment in milliseconds.  
62 - (default: 300)  
63 -  
64 - -tailmargin msec  
65 - Silence margin at the end of speech segment in milliseconds.  
66 - (default: 400)  
67 -  
68 - -zmean  
69 - This option enables DC offset removal.  
70 -  
71 - -smpFreq Hz  
72 - Set sampling rate in Hz. (default: 16,000)  
73 -  
74 - -48  
75 - Record input with 48kHz sampling, and down-sample it to 16kHz  
76 - on-the-fly. This option is valid for 16kHz model only. The  
77 - down-sampling routine was ported from sptk. (Rev. 4.0)  
78 -  
79 - -NA devicename  
80 - Host name for DatLink server input (-input netaudio).  
81 -  
82 - -adport port_number  
83 - With -input adinnet, specify adinnet port number to listen.  
84 - (default: 5530)  
85 -  
86 - -nostrip  
87 - Julius by default removes successive zero samples in input speech  
88 - data. This option stop it.  
89 -  
90 - -C jconffile  
91 - Load a jconf file at here. The content of the jconffile will be  
92 - expanded at this point.  
93 -  
94 - -plugindir dirlist  
95 - Specify which directories to load plugin. If several direcotries  
96 - exist, specify them by colon-separated list.  
97 -  
98 -ENVIRONMENT VARIABLES  
99 - ALSADEV  
100 - Device name string for ALSA. (default: "default")  
101 -  
102 - AUDIODEV  
103 - Device name string for OSS. (default: "/dev/dsp")  
104 -  
105 - LATENCY_MSEC  
106 - Input latency of microphone input in milliseconds. Smaller value  
107 - will shorten latency but sometimes make process unstable. Default  
108 - value will depend on the running OS.  
109 -  
110 -SEE ALSO  
111 - julius ( 1 ) ,  
112 - adintool ( 1 )  
113 -  
114 -COPYRIGHT  
115 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
116 -  
117 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
118 -  
119 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
120 - Technology  
121 -  
122 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
123 - Technology  
124 -  
125 -LICENSE  
126 - The same as Julius.  
127 -  
128 -  
129 -  
130 - 10/02/2008 ADINREC(1)  
recognize/src/julius/doc/manuals/adintool.txt
View file @e1cf2b2
@@ -1,214 +0,0 @@ @@ -1,214 +0,0 @@
1 - adintool  
2 -  
3 -ADINTOOL(1) ADINTOOL(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - adintool  
9 - - a tool to record / split / send / receive audio streams  
10 -  
11 -SYNOPSIS  
12 - adintool {-in inputdev} {-out outputdev} [options...]  
13 -  
14 -DESCRIPTION  
15 - adintool analyzes speech input, finds speech segments skipping silence,  
16 - and records the detected segments in various ways. It performs speech  
17 - detection based on zerocross number and power (level), and records the  
18 - detected parts to files or other output devices sucessively.  
19 -  
20 -  
21 - adintool is a upper version of adinrec with various functions.  
22 - Supported input device are: microphone input, a speech file, standard  
23 - tty input, and network socket (called adin-net server mode). Julius  
24 - plugin can be also used. Detected speech segments will be saved to  
25 - output devices: speech files, standard tty output, and network socket  
26 - (called adin-net client mode). For example, you can split the incoming  
27 - speech to segments and send them to Julius to be recognized.  
28 -  
29 - Output format is WAV, 16bit (signed short), monoral. If the file  
30 - already exist, it will be overridden.  
31 -  
32 -OPTIONS  
33 - All Julius options can be set. Only audio input related options are  
34 - treated and others are silently skipped. Below is a list of options.  
35 -  
36 - adintool specific options  
37 - -freq Hz  
38 - Set sampling rate in Hz. (default: 16,000)  
39 -  
40 - -in inputdev  
41 - Audio input device. "mic" to capture via microphone input, "file"  
42 - for audio file input, and "stdin" to read raw data from  
43 - standard-input. For file input, file name prompt will appear after  
44 - startup. Use "adinnet" to make adintool as "adinnet server",  
45 - receiving data from client via network socket. Default port number  
46 - is 5530, which can be altered by option "-inport".  
47 -  
48 - Alternatively, input device can be set by "-input" option, in which  
49 - case you can use plugin input.  
50 -  
51 - -out outputdev  
52 - Audio output device store the data. Specify "file" to save to file,  
53 - in which the output filename should be given by "-filename". Use  
54 - "stdout" to standard out. "adinnet" will make adintool to be an  
55 - adinnet client, sending speech data to a server via tcp/ip socket.  
56 - When using "adinnet" output, the server name to send data should be  
57 - specified by "-server". The default port number is 5530, which can  
58 - be changed by "-port" option.  
59 -  
60 - -inport num  
61 - When adintool becomes adinnet server to receive data (-in adinnet),  
62 - set the port number to listen. (default: 5530)  
63 -  
64 - -server [host] [,host...]  
65 - When output to adinnet server (-out adinnet), set the hostname. You  
66 - can send to multiple hosts by specifying their hostnames as  
67 - comma-delimited list like "host1,host2,host3".  
68 -  
69 - -port [num] [,num...]  
70 - When adintool send a data to adinnet server (-out adinnet), set the  
71 - port number to connect. (default: 5530) For multiple servers,  
72 - specify port numbers for all servers like "5530,5530,5531".  
73 -  
74 - -filename file  
75 - When output to file (-out file), set the output filename. The actual  
76 - file name will be as "file.0000.wav" , "file.0001.wav" and so on,  
77 - where the four digit number increases as speech segment detected.  
78 - The initial number will be set to 0 by default, which can be changed  
79 - by "-startid" option. When using "-oneshot" option to save only the  
80 - first segment, the input will be saved as "file".  
81 -  
82 - -startid number  
83 - At file output, set the initial file number. (default: 0)  
84 -  
85 - -oneshot  
86 - Exit after the end of first speech segment.  
87 -  
88 - -nosegment  
89 - Do not perform speech detection for input, just treat all the input  
90 - as a single valid segment.  
91 -  
92 - -raw  
93 - Output as RAW file (no header).  
94 -  
95 - -autopause  
96 - When output to adinnet server, adintool enter pause state at every  
97 - end of speech segment. It will restart when the destination adinnet  
98 - server sends it a resume signal.  
99 -  
100 - -loosesync  
101 - When output to multiple adinnet server, not to do strict  
102 - synchronization for restart. By default, when adintool has entered  
103 - pause state, it will not restart until resume commands are received  
104 - from all servers. This option will allow restart at least one  
105 - restart command has arrived.  
106 -  
107 - -rewind msec  
108 - When input is a live microphone device, and there has been some  
109 - continuing input at the moment adintool resumes, it start recording  
110 - backtracking by the specified milliseconds.  
111 -  
112 - Concerning Julius options  
113 - -input {mic|rawfile|adinnet|stdin|netaudio|esd|alsa|oss}  
114 - Choose speech input source. Specify 'file' or 'rawfile' for waveform  
115 - file. On file input, users will be prompted to enter the file name  
116 - from stdin.  
117 -  
118 - 'mic' is to get audio input from a default live microphone device,  
119 - and 'adinnet' means receiving waveform data via tcpip network from  
120 - an adinnet client. 'netaudio' is from DatLink/NetAudio input, and  
121 - 'stdin' means data input from standard input.  
122 -  
123 - At Linux, you can choose API at run time by specifying alsa, oss and  
124 - esd.  
125 -  
126 - -lv thres  
127 - Level threshold for speech input detection. Values should be in  
128 - range from 0 to 32767. (default: 2000)  
129 -  
130 - -zc thres  
131 - Zero crossing threshold per second. Only input that goes over the  
132 - level threshold (-lv) will be counted. (default: 60)  
133 -  
134 - -headmargin msec  
135 - Silence margin at the start of speech segment in milliseconds.  
136 - (default: 300)  
137 -  
138 - -tailmargin msec  
139 - Silence margin at the end of speech segment in milliseconds.  
140 - (default: 400)  
141 -  
142 - -zmean  
143 - This option enables DC offset removal.  
144 -  
145 - -smpFreq Hz  
146 - Set sampling rate in Hz. (default: 16,000)  
147 -  
148 - -48  
149 - Record input with 48kHz sampling, and down-sample it to 16kHz  
150 - on-the-fly. This option is valid for 16kHz model only. The  
151 - down-sampling routine was ported from sptk. (Rev. 4.0)  
152 -  
153 - -NA devicename  
154 - Host name for DatLink server input (-input netaudio).  
155 -  
156 - -adport port_number  
157 - With -input adinnet, specify adinnet port number to listen.  
158 - (default: 5530)  
159 -  
160 - -nostrip  
161 - Julius by default removes successive zero samples in input speech  
162 - data. This option stop it.  
163 -  
164 - -C jconffile  
165 - Load a jconf file at here. The content of the jconffile will be  
166 - expanded at this point.  
167 -  
168 - -plugindir dirlist  
169 - Specify which directories to load plugin. If several direcotries  
170 - exist, specify them by colon-separated list.  
171 -  
172 -ENVIRONMENT VARIABLES  
173 - ALSADEV  
174 - (using mic input with alsa device) specify a capture device name. If  
175 - not specified, "default" will be used.  
176 -  
177 - AUDIODEV  
178 - (using mic input with oss device) specify a capture device path. If  
179 - not specified, "/dev/dsp" will be used.  
180 -  
181 - LATENCY_MSEC  
182 - Try to set input latency of microphone input in milliseconds.  
183 - Smaller value will shorten latency but sometimes make process  
184 - unstable. Default value will depend on the running OS.  
185 -  
186 -EXAMPLES  
187 - Record microphone input to files: "data.0000.wav", "data.0001.wav" and  
188 - so on:  
189 - Split a long speech file "foobar.raw" into "foobar.1500.wav",  
190 - "foobar.1501.wav" ...:  
191 - Copy an entire audio file via network socket.  
192 - Detect speech segment, send to Julius via network and recognize it:  
193 -  
194 -SEE ALSO  
195 - julius ( 1 ) ,  
196 - adinrec ( 1 )  
197 -  
198 -COPYRIGHT  
199 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
200 -  
201 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
202 -  
203 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
204 - Technology  
205 -  
206 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
207 - Technology  
208 -  
209 -LICENSE  
210 - The same as Julius.  
211 -  
212 -  
213 -  
214 - 10/02/2008 ADINTOOL(1)  
recognize/src/julius/doc/manuals/dfa_determinize.txt
View file @e1cf2b2
@@ -1,51 +0,0 @@ @@ -1,51 +0,0 @@
1 - dfa_determinize  
2 -  
3 -DFA_DETERMINIZE(1) DFA_DETERMINIZE(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - dfa_determinize  
9 - - Determinize NFA grammar network.  
10 -  
11 -SYNOPSIS  
12 - dfa_determinize [-o outfile] {dfafile}  
13 -  
14 -DESCRIPTION  
15 - dfa_determinize converts a non-deterministic .dfa file into  
16 - deterministic DFA. Output to standard output, or file specified by "-o"  
17 - option.  
18 -  
19 - This additional tool is not necessary on a grammar building procedure  
20 - in Julius, since the grammar network generated by mkdfa.pl is always  
21 - determinized.  
22 -  
23 -OPTIONS  
24 - -o outfile  
25 - Outout file. If not specified, output to stdout.  
26 -  
27 -EXAMPLES  
28 - Determinize foo.dfa to bar.dfa:  
29 - Another way:  
30 -  
31 -SEE ALSO  
32 - mkdfa.pl ( 1 ) ,  
33 - dfa_minimize ( 1 )  
34 -  
35 -COPYRIGHT  
36 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
37 -  
38 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
39 -  
40 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
41 - Technology  
42 -  
43 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
44 - Technology  
45 -  
46 -LICENSE  
47 - The same as Julius.  
48 -  
49 -  
50 -  
51 - 10/02/2008 DFA_DETERMINIZE(1)  
recognize/src/julius/doc/manuals/dfa_minimize.txt
View file @e1cf2b2
@@ -1,49 +0,0 @@ @@ -1,49 +0,0 @@
1 - dfa_minimize  
2 -  
3 -DFA_MINIMIZE(1) DFA_MINIMIZE(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - dfa_minimize  
9 - - Minimize a DFA grammar network  
10 -  
11 -SYNOPSIS  
12 - dfa_minimize [-o outfile] {dfafile}  
13 -  
14 -DESCRIPTION  
15 - dfa_minimize will convert an .dfa file to an equivalent minimal form.  
16 - Output to standard output, or to a file specified by "-o" option.  
17 -  
18 - On version 3.5.3 and later, mkdfa.pl invokes this tool inside, and the  
19 - output .dfa file will be always minimized, so you do not need to use  
20 - this manually.  
21 -  
22 -OPTIONS  
23 - -o outfile  
24 - Output file. If not specified output to standard output.  
25 -  
26 -EXAMPLES  
27 - Minimize foo.dfa to bar.dfa:  
28 - Another way:  
29 -  
30 -SEE ALSO  
31 - mkdfa.pl ( 1 )  
32 -  
33 -COPYRIGHT  
34 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
35 -  
36 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
37 -  
38 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
39 - Technology  
40 -  
41 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
42 - Technology  
43 -  
44 -LICENSE  
45 - The same as Julius.  
46 -  
47 -  
48 -  
49 - 10/02/2008 DFA_MINIMIZE(1)  
recognize/src/julius/doc/manuals/generate-ngram.txt
View file @e1cf2b2
@@ -1,62 +0,0 @@ @@ -1,62 +0,0 @@
1 - generate-ngram  
2 -  
3 -GENERATE-NGRAM(1) GENERATE-NGRAM(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - generate-ngram  
9 - - random sentence generator from N-gram  
10 -  
11 -SYNOPSIS  
12 - generate-ngram [options...] {binary_ngram}  
13 -  
14 -DESCRIPTION  
15 - generate-ngram is a tool to generate sentences randomly according to  
16 - the given N-gram language model. The N-gram model file binary_ngram  
17 - should be an binary format.  
18 -  
19 -OPTIONS  
20 - -n num  
21 - Number of sentences to generate (default: 10)  
22 -  
23 - -N  
24 - Specify which length of N-gram to use (default: available max in the  
25 - given model)  
26 -  
27 - -bos  
28 - Beginning-of-sentence word (default: "<s>")  
29 -  
30 - -eos  
31 - End-of-sentence word (default: "</s>")  
32 -  
33 - -ignore  
34 - Specify a word to be supressed from output (default: "<UNK")  
35 -  
36 - -v  
37 - Verbose output.  
38 -  
39 - -debug  
40 - Debug output.  
41 -  
42 -SEE ALSO  
43 - julius ( 1 ) ,  
44 - mkbingram ( 1 )  
45 -  
46 -COPYRIGHT  
47 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
48 -  
49 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
50 -  
51 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
52 - Technology  
53 -  
54 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
55 - Technology  
56 -  
57 -LICENSE  
58 - The same as Julius.  
59 -  
60 -  
61 -  
62 - 10/02/2008 GENERATE-NGRAM(1)  
recognize/src/julius/doc/manuals/generate.txt
View file @e1cf2b2
@@ -1,76 +0,0 @@ @@ -1,76 +0,0 @@
1 - generate  
2 -  
3 -GENERATE(1) GENERATE(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - generate  
9 - - random sentence generator from a grammar  
10 -  
11 -SYNOPSIS  
12 - generate [-v] [-t] [-n num] [-s spname] {prefix}  
13 -  
14 -DESCRIPTION  
15 - This small program randomly generates sentences that are acceptable by  
16 - the given grammar.  
17 -  
18 -  
19 - .dfa, .dict and .term files are needed to execute. They can be  
20 - generated from .grammar and .voca file by mkdfa.pl.  
21 -  
22 -OPTIONS  
23 - -t  
24 - Output in word's category name.  
25 -  
26 - -n num  
27 - Set number of sentences to be generated (default: 10)  
28 -  
29 - -s spname  
30 - the name string of short-pause word to be supressed (default: "sp")  
31 -  
32 - -v  
33 - Debug output mode.  
34 -  
35 -EXAMPLES  
36 - Exmple output of a sample grammar "fruit":  
37 -  
38 - % generate fruit  
39 - Stat: init_voca: read 36 words  
40 - Reading in term file (optional)...done  
41 - 15 categories, 36 words  
42 - DFA has 26 nodes and 42 arcs  
43 - -----  
44 - <s> I WANT ONE APPLE </s>  
45 - <s> I WANT TEN PEARS </s>  
46 - <s> CAN I HAVE A PINEAPPLE </s>  
47 - <s> I WANT ONE PEAR </s>  
48 - <s> COULD I HAVE A BANANA </s>  
49 - <s> I WANT ONE APPLE PLEASE </s>  
50 - <s> I WANT NINE APPLES </s>  
51 - <s> NINE APPLES </s>  
52 - <s> I WANT ONE PINEAPPLE </s>  
53 - <s> I WANT A PEAR </s>  
54 -  
55 -  
56 -SEE ALSO  
57 - mkdfa.pl ( 1 ) ,  
58 - generate-ngram ( 1 )  
59 -  
60 -COPYRIGHT  
61 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
62 -  
63 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
64 -  
65 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
66 - Technology  
67 -  
68 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
69 - Technology  
70 -  
71 -LICENSE  
72 - The same as Julius.  
73 -  
74 -  
75 -  
76 - 10/02/2008 GENERATE(1)  
recognize/src/julius/doc/manuals/gram2sapixml.pl.txt
View file @e1cf2b2
@@ -1,47 +0,0 @@ @@ -1,47 +0,0 @@
1 - gram2sapixml.pl  
2 -  
3 -GRAM2SAPIXML.PL(1) GRAM2SAPIXML.PL(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - gram2sapixml.pl  
9 - - convert Julius grammar to SAPI XML grammar format  
10 -  
11 -SYNOPSIS  
12 - gram2sapixml.pl [prefix...]  
13 -  
14 -DESCRIPTION  
15 - gram2sapixml.pl converts a recognition grammar file of Julius  
16 - (.grammar, .voca) to Microsoft SAPI XML grammar format. prefix should  
17 - be a file name of target grammar, excluding suffixes. If multiple  
18 - argument is given, each will be process sequencialy in turn.  
19 -  
20 - The internal character set should be in UTF-8 format. By default  
21 - gram2sapixml.pl assume input in EUC-JP encoding and tries to convert it  
22 - to UTF-8 using iconv. You may want to disable this feature within the  
23 - script.  
24 -  
25 - It will fail to convert a left recursive rule in the grammar. When  
26 - fails, it will leave the source rules in the target .xml file, so you  
27 - should modify the output manually to solve it.  
28 -  
29 -SEE ALSO  
30 - mkdfa.pl ( 1 )  
31 -  
32 -DIAGNOSTICS  
33 - The conversion procedure is somewhat dumb one, only converting the  
34 - non-terminal symbols and terminal symbols (=word category name) into  
35 - corresponding rules one by one. This is only a help tool, and you will  
36 - need a manual inspection and editing to use it on a real SAPI  
37 - application.  
38 -  
39 -COPYRIGHT  
40 - Copyright (c) 2002 Takashi Sumiyoshi  
41 -  
42 -LICENSE  
43 - The same as Julius.  
44 -  
45 -  
46 -  
47 - 10/02/2008 GRAM2SAPIXML.PL(1)  
recognize/src/julius/doc/manuals/jclient.pl.txt
View file @e1cf2b2
@@ -1,46 +0,0 @@ @@ -1,46 +0,0 @@
1 - jclient.pl  
2 -  
3 -JCLIENT.PL(1) JCLIENT.PL(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - jclient.pl  
9 - - sample client for module mode (perl version)  
10 -  
11 -SYNOPSIS  
12 - jclient.pl  
13 -  
14 -DESCRIPTION  
15 - This is yet another sample client written in perl. It will connect to  
16 - Julius running in module mode, receive recognition results from Julius,  
17 - and cna send commands to control Julius.  
18 -  
19 - This is a tiny program with only 57 lines. You can use it for free.  
20 -  
21 -EXAMPLES  
22 - Invoke Julius with module mode by specifying "-module" option:  
23 - Then, at other terminal or other host, invoke jclient.pl like below.  
24 - The default hostname is "localhost", and port number is 10500. You can  
25 - change them by editing the top part of the script.  
26 - It will then receive the outputs of Julius and output the raw message  
27 - to standard out. Also, by inputting a raw module command to the  
28 - standard input of jclient.pl, it will be sent to Julius. See manuals  
29 - for the specification of module mode.  
30 -  
31 -SEE ALSO  
32 - julius ( 1 ) ,  
33 - jcontrol ( 1 )  
34 -  
35 -COPYRIGHT  
36 - "jclient.pl" has been developed by Dr. Ryuichi Nisimura  
37 - (nisimura@sys.wakayama-u.ac.jp). Use at your own risk.  
38 -  
39 - If you have any feedback, comment or request, please contact the E-mail  
40 - address above, or look at the Web page below.  
41 -  
42 - http://w3voice.jp/  
43 -  
44 -  
45 -  
46 - 10/02/2008 JCLIENT.PL(1)  
recognize/src/julius/doc/manuals/jcontrol.txt
View file @e1cf2b2
@@ -1,173 +0,0 @@ @@ -1,173 +0,0 @@
1 - jcontrol  
2 -  
3 -JCONTROL(1) JCONTROL(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - jcontrol  
9 - - a sample module client written in C  
10 -  
11 -SYNOPSIS  
12 - jcontrol {hostname} [portnum]  
13 -  
14 -DESCRIPTION  
15 - jcontrol is a simple console program to control julius running on other  
16 - host via network API. It can send command to Julius, and receive  
17 - messages from Julius.  
18 -  
19 - When invoked, jcontrol tries to connect to Julius running in "module  
20 - mode" on specified hostname. After connection established, jcontrol  
21 - waits for user commands from standard input.  
22 -  
23 - When user types a command to jcontrol, it will be interpreted and cor-  
24 - responding API command will be sent to Julius. When a message is  
25 - received from Julius, its content will be output to standard output.  
26 -  
27 - For the details about the API, see the related documents.  
28 -  
29 -OPTIONS  
30 - hostname  
31 - Host name where Julius is runnning in module mode.  
32 -  
33 - portnum  
34 - port number (default: 10500)  
35 -  
36 -COMMANDS  
37 - jcontrol interprets commands from standard input. Below is a list of  
38 - all commands.  
39 -  
40 - Engine control  
41 - pause  
42 - Stop Julius and enter into paused status. In paused status, Julius  
43 - will not run recognition even if speech input occurs. When this  
44 - command is issued while recognition is running, Julius will stop  
45 - after the recognition has been finished.  
46 -  
47 - terminate  
48 - Same as pause, but discard the current speech input when received  
49 - command in the middle of recognition process.  
50 -  
51 - resume  
52 - Restart Julius that has been paused or terminated.  
53 -  
54 - inputparam arg  
55 - Tell Julius how to deal with speech input in case grammar is changed  
56 - just when recognition is running. Specify one: "TERMINATE", "PAUSE"  
57 - or "WAIT".  
58 -  
59 - version  
60 - Tell Julius to send version description string.  
61 -  
62 - status  
63 - Tell Julius to send the system status (active / sleep)  
64 -  
65 - Grammar handling  
66 - changegram prefix  
67 - Send a new grammar "prefix.dfa" and "prefix.dict", and tell julius  
68 - to use it as a new grammar. All the current grammars used in the  
69 - current process of Julius will be deleted and replaced to the  
70 - specifed grammar.  
71 -  
72 - addgram prefix  
73 - Send a new grammar "prefix.dfa" and "prefix.dict" and add it to the  
74 - current grammar.  
75 -  
76 - deletegram gramlist  
77 - Tell Julius to delete existing grammar. The grammar can be specified  
78 - by either prefix name or number ID. The number ID can be determined  
79 - from the message sent from Julius at each time grammar information  
80 - has changed. When want to delete more than one grammar, specify all  
81 - of them as comma-sparated.  
82 -  
83 - deactivategram gramlist  
84 - Tell Julius to de-activate a specified grammar. The specified  
85 - grammar will still be kept but will not be used for recognition.  
86 -  
87 - The target grammar can be specified by either prefix name or number  
88 - ID. The number ID can be determined from the message sent from  
89 - Julius at each time grammar information has changed. When want to  
90 - delete more than one grammar, specify all of them as comma-sparated.  
91 -  
92 - activategram gramlist  
93 - Tell Julius to activate previously de-activated grammar. The target  
94 - grammar can be specified by either prefix name or number ID. The  
95 - number ID can be determined from the message sent from Julius at  
96 - each time grammar information has changed. When want to delete more  
97 - than one grammar, specify all of them as comma-sparated.  
98 -  
99 - addword grammar_name_or_id dictfile  
100 - Add the recognition word entries in the specified dictfile to the  
101 - specified grammar on current process.  
102 -  
103 - syncgram  
104 - Force synchronize grammar status, like unix command "sync".  
105 -  
106 - Process management  
107 - Julius-4 supports multi-model recognition nad multi decoding. In this  
108 - case it is possible to control each recognition process, as defined by  
109 - "-SR" option, from module client.  
110 -  
111 - In multi decoding mode, the module client holds "current process", and  
112 - the process commands and grammar related commands will be issued toward  
113 - the current process.  
114 -  
115 - listprocess  
116 - Tell Julius to send the list of existing recognition process.  
117 -  
118 - currentprocess procname  
119 - Switch the current process to the process specified by the name.  
120 -  
121 - shiftprocess  
122 - Rotate the current process. At each call the current process will be  
123 - changed to the next one.  
124 -  
125 - addprocess jconffile  
126 - Tell Julisu to load a new recognition process into engine. The  
127 - argument jconffile should be a jconf file that contains only one set  
128 - of LM options and one SR definition. Note that the file should be  
129 - visible on the running Julius, since jcontrol only send the path  
130 - name and Julius actually read the jconf file.  
131 -  
132 - The new LM and SR process will have the name of the jconffile.  
133 -  
134 - delprocess procname  
135 - Delete the specified recognition process from the engine.  
136 -  
137 - deactivateprocess procname  
138 - Tell Julius to temporary stop the specified recognition process. The  
139 - stopped process will not be executed for the input until activated  
140 - again.  
141 -  
142 - activateprocess procname  
143 - Tell Julius to activate the temporarily stopped process.  
144 -  
145 -EXAMPLES  
146 - The dump messages from Julius are output to tty with prefix ">"  
147 - appended to each line. Julius can be started in module mode like this:  
148 - jcontrolcan be launched with the host name:  
149 - It will then receive the outputs of Julius and output the raw message  
150 - to standard out. Also, by inputting the commands above to the standard  
151 - input of jcontrol, it will be sent to Julius. See manuals for the  
152 - specification of module mode.  
153 -  
154 -SEE ALSO  
155 - julius ( 1 )  
156 -  
157 -COPYRIGHT  
158 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
159 -  
160 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
161 -  
162 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
163 - Technology  
164 -  
165 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
166 - Technology  
167 -  
168 -LICENSE  
169 - The same as Julius.  
170 -  
171 -  
172 -  
173 - 10/02/2008 JCONTROL(1)  
recognize/src/julius/doc/manuals/julius.txt
View file @e1cf2b2
@@ -1,1287 +0,0 @@ @@ -1,1287 +0,0 @@
1 - julius  
2 -  
3 -JULIUS(1) JULIUS(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - julius  
9 - - open source multi-purpose LVCSR engine  
10 -  
11 -SYNOPSIS  
12 - julius [-C jconffile] [options...]  
13 -  
14 -DESCRIPTION  
15 - julius is a high-performance, multi-purpose, open-source speech  
16 - recognition engine for researchers and developers. It is capable of  
17 - performing almost real-time recognition of continuous speech with over  
18 - 60k-word 3-gram language model and triphone HMM model, on most current  
19 - PCs. julius can perform recognition on audio files, live microphone  
20 - input, network input and feature parameter files.  
21 -  
22 - The core recognition module is implemented as C library called  
23 - "JuliusLib". It can also be extended by plug-in facility.  
24 -  
25 - Supported Models  
26 - julius needs a language model and an acoustic model to run as a speech  
27 - recognizer. julius supports the following models.  
28 -  
29 - Acoustic model  
30 - Sub-word HMM (Hidden Markov Model) in HTK ascii format are  
31 - supported. Phoneme models (monophone), context dependent phoneme  
32 - models (triphone), tied-mixture and phonetic tied-mixture models  
33 - of any unit can be used. When using context dependent models,  
34 - inter-word context dependency is also handled. Multi-stream  
35 - feature and MSD-HMM is also supported. You can further use a  
36 - tool mkbinhmm to convert the ascii HMM file to a compact binary  
37 - format for faster loading.  
38 -  
39 - Note that julius itself can only extract MFCC features from  
40 - speech data. If you use acoustic HMM trained for other feature,  
41 - you should give the input in HTK parameter file of the same  
42 - feature type.  
43 -  
44 - Language model: word N-gram  
45 - Word N-gram language model, up to 10-gram, is supported. Julius  
46 - uses different N-gram for each pass: left-to-right 2-gram on 1st  
47 - pass, and right-to-left N-gram on 2nd pass. It is recommended to  
48 - use both LR 2-gram and RL N-gram for Julius. However, you can  
49 - use only single LR N-gram or RL N-gram. In such case,  
50 - approximated LR 2-gram computed from the given N-gram will be  
51 - applied at the first pass.  
52 -  
53 - The Standard ARPA format is supported. In addition, a binary  
54 - format is also supported for efficiency. The tool mkbingram(1)  
55 - can convert ARPA format N-gram to binary format.  
56 -  
57 - Language model: grammar  
58 - The grammar format is an original one, and tools to create a  
59 - recognirion grammar are included in the distribution. A grammar  
60 - consists of two files: one is a 'grammar' file that describes  
61 - sentence structures in a BNF style, using word 'category' name  
62 - as terminate symbols. Another is a 'voca' file that defines  
63 - words with its pronunciations (i.e. phoneme sequences) for each  
64 - category. They should be converted by mkdfa.pl(1) to a  
65 - deterministic finite automaton file (.dfa) and a dictionary file  
66 - (.dict), respectively. You can also use multiple grammars.  
67 -  
68 - Language model: isolated word  
69 - You can perform isolated word recognition using only word  
70 - dictionary. With this model type, Julius will perform rapid one  
71 - pass recognition with static context handling. Silence models  
72 - will be added at both head and tail of each word. You can also  
73 - use multiple dictionaries in a process.  
74 -  
75 - Search Algorithm  
76 - Recognition algorithm of julius is based on a two-pass strategy. Word  
77 - 2-gram and reverse word 3-gram is used on the respective passes. The  
78 - entire input is processed on the first pass, and again the final  
79 - searching process is performed again for the input, using the result of  
80 - the first pass to narrow the search space. Specifically, the  
81 - recognition algorithm is based on a tree-trellis heuristic search  
82 - combined with left-to-right frame-synchronous beam search and  
83 - right-to-left stack decoding search.  
84 -  
85 - When using context dependent phones (triphones), interword contexts are  
86 - taken into consideration. For tied-mixture and phonetic tied-mixture  
87 - models, high-speed acoustic likelihood calculation is possible using  
88 - gaussian pruning.  
89 -  
90 - For more details, see the related documents.  
91 -  
92 -OPTIONS  
93 - These options specify the models, system behaviors and various search  
94 - parameters to Julius. These option can be set at the command line, but  
95 - it is recommended that you write them in a text file as a "jconf file",  
96 - and specify it by "-C" option.  
97 -  
98 - Applications incorporating JuliusLib also use these options to set the  
99 - parameters of core recognition engine. For example, a jconf file can be  
100 - loaded to the enine by calling j_config_load_file_new() with the jconf  
101 - file name as argument.  
102 -  
103 - Please note that relative paths in a jconf file should be relative to  
104 - the jconf file itself, not the current working directory.  
105 -  
106 - Below are the details of all options, gathered by group.  
107 -  
108 - Julius application option  
109 - These are application options of Julius, outside of JuliusLib. It  
110 - contains parameters and switches for result output, character set  
111 - conversion, log level, and module mode options. These option are  
112 - specific to Julius, and cannot be used at applications using JuliusLib  
113 - other than Julius.  
114 -  
115 - -outfile  
116 - On file input, this option write the recognition result of each file  
117 - to a separate file. The output file of an input file will be the  
118 - same name but the suffix will be changed to ".out". (rev.4.0)  
119 -  
120 - -separatescore  
121 - Output the language and acoustic scores separately.  
122 -  
123 - -callbackdebug  
124 - Print the callback names at each call for debug. (rev.4.0)  
125 -  
126 - -charconv from to  
127 - Print with character set conversion. from is the source character  
128 - set used in the language model, and to is the target character set  
129 - you want to get.  
130 -  
131 - On Linux, the arguments should be a code name. You can obtain the  
132 - list of available code names by invoking the command "iconv --list".  
133 - On Windows, the arguments should be a code name or codepage number.  
134 - Code name should be one of "ansi", "mac", "oem", "utf-7", "utf-8",  
135 - "sjis", "euc". Or you can specify any codepage number supported at  
136 - your environment.  
137 -  
138 - -nocharconv  
139 - Disable character conversion.  
140 -  
141 - -module [port]  
142 - Run Julius on "Server Module Mode". After startup, Julius waits for  
143 - tcp/ip connection from client. Once connection is established,  
144 - Julius start communication with the client to process incoming  
145 - commands from the client, or to output recognition results, input  
146 - trigger information and other system status to the client. The  
147 - default port number is 10500.  
148 -  
149 - -record dir  
150 - Auto-save all input speech data into the specified directory. Each  
151 - segmented inputs are recorded each by one. The file name of the  
152 - recorded data is generated from system time when the input ends, in  
153 - a style of YYYY.MMDD.HHMMSS.wav. File format is 16bit monoral WAV.  
154 - Invalid for mfcfile input.  
155 -  
156 - With input rejection by -rejectshort, the rejected input will also  
157 - be recorded even if they are rejected.  
158 -  
159 - -logfile file  
160 - Save all log output to a file instead of standard output. (Rev.4.0)  
161 -  
162 - -nolog  
163 - Disable all log output. (Rev.4.0)  
164 -  
165 - -help  
166 - Output help message and exit.  
167 -  
168 - Global options  
169 - These are model-/search-dependent options relating audio input, sound  
170 - detection, GMM, decoding algorithm, plugin facility, and others. Global  
171 - options should be placed before any instance declaration (-AM, -LM, or  
172 - -SR), or just after "-GLOBAL" option.  
173 -  
174 - Audio input  
175 - -input  
176 - {mic|rawfile|mfcfile|adinnet|stdin|netaudio|alsa|oss|esd}  
177 - Choose speech input source. Specify 'file' or 'rawfile' for  
178 - waveform file, 'htkparam' or 'mfcfile' for HTK parameter  
179 - file. On file input, users will be prompted to enter the file  
180 - name from stdin, or you can use -filelist option to specify  
181 - list of files to process.  
182 -  
183 - 'mic' is to get audio input from a default live microphone  
184 - device, and 'adinnet' means receiving waveform data via tcpip  
185 - network from an adinnet client. 'netaudio' is from  
186 - DatLink/NetAudio input, and 'stdin' means data input from  
187 - standard input.  
188 -  
189 - For waveform file input, only WAV (no compression) and RAW  
190 - (noheader, 16bit, big endian) are supported by default. Other  
191 - format can be read when compiled with libsnd library. To see  
192 - what format is actually supported, see the help message using  
193 - option -help. For stdin input, only WAV and RAW is supported.  
194 - (default: mfcfile)  
195 -  
196 - At Linux, you can choose API at run time by specifying alsa,  
197 - oss and esd.  
198 -  
199 - -filelist filename  
200 - (With -input rawfile|mfcfile) perform recognition on all  
201 - files listed in the file. The file should contain input file  
202 - per line. Engine will end when all of the files are  
203 - processed.  
204 -  
205 - -notypecheck  
206 - By default, Julius checks the input parameter type whether it  
207 - matches the AM or not. This option will disable the check and  
208 - force engine to use the input vector as is.  
209 -  
210 - -48  
211 - Record input with 48kHz sampling, and down-sample it to 16kHz  
212 - on-the-fly. This option is valid for 16kHz model only. The  
213 - down-sampling routine was ported from sptk. (Rev. 4.0)  
214 -  
215 - -NA devicename  
216 - Host name for DatLink server input (-input netaudio).  
217 -  
218 - -adport port_number  
219 - With -input adinnet, specify adinnet port number to listen.  
220 - (default: 5530)  
221 -  
222 - -nostrip  
223 - Julius by default removes successive zero samples in input  
224 - speech data. This option inhibits the removal.  
225 -  
226 - -zmean , -nozmean  
227 - This option enables/disables DC offset removal of input  
228 - waveform. Offset will be estimated from the whole input. For  
229 - microphone / network input, zero mean of the first 48000  
230 - samples (3 seconds in 16kHz sampling) will be used for the  
231 - estimation. (default: disabled)  
232 -  
233 - This option uses static offset for the channel. See also  
234 - -zmeansource for frame-wise offset removal.  
235 -  
236 - Speech detection by level and zero-cross  
237 - -cutsilence , -nocutsilence  
238 - Turn on / off the speech detection by level and zero-cross.  
239 - Default is on for mic / adinnet input, and off for files.  
240 -  
241 - -lv thres  
242 - Level threshold for speech input detection. Values should be  
243 - in range from 0 to 32767. (default: 2000)  
244 -  
245 - -zc thres  
246 - Zero crossing threshold per second. Only input that goes over  
247 - the level threshold (-lv) will be counted. (default: 60)  
248 -  
249 - -headmargin msec  
250 - Silence margin at the start of speech segment in  
251 - milliseconds. (default: 300)  
252 -  
253 - -tailmargin msec  
254 - Silence margin at the end of speech segment in milliseconds.  
255 - (default: 400)  
256 -  
257 - Input rejection  
258 - Two simple front-end input rejection methods are implemented,  
259 - based on input length and average power of detected segment. The  
260 - rejection by average power is experimental, and can be enabled  
261 - by --enable-power-reject on compilation. Valid for MFCC feature  
262 - with power coefficient and real-time input only.  
263 -  
264 - For GMM-based input rejection see the GMM section below.  
265 -  
266 - -rejectshort msec  
267 - Reject input shorter than specified milliseconds. Search will  
268 - be terminated and no result will be output.  
269 -  
270 - -powerthres thres  
271 - Reject the inputted segment by its average energy. If the  
272 - average energy of the last recognized input is below the  
273 - threshold, Julius will reject the input. (Rev.4.0)  
274 -  
275 - This option is valid when --enable-power-reject is specified  
276 - at compilation time.  
277 -  
278 - Gaussian mixture model / GMM-VAD  
279 - GMM will be used for input rejection by accumulated score, or  
280 - for front-end GMM-based VAD when --enable-gmm-vad is specified.  
281 -  
282 - NOTE: You should also set the proper MFCC parameters required  
283 - for the GMM, specifying the acoustic parameters described in AM  
284 - section -AM_GMM.  
285 -  
286 - When GMM-based VAD is enabled, the voice activity score will be  
287 - calculated at each frame as front-end processing. The value will  
288 - be computed as \[ \max_{m \in M_v} p(x|m) - \max_{m \in M_n}  
289 - p(x|m) \] where $M_v$ is a set of voice GMM, and $M_n$ is a set  
290 - of noise GMM whose names should be specified by -gmmreject. The  
291 - activity score will be then averaged for the last N frames,  
292 - where N is specified by -gmmmargin. Julius updates the averaged  
293 - activity score at each frame, and detect speech up-trigger when  
294 - the value gets higher than a value specified by -gmmup, and  
295 - detecgt down-trigger when it gets lower than a value of  
296 - -gmmdown.  
297 -  
298 - -gmm hmmdefs_file  
299 - GMM definition file in HTK format. If specified, GMM-based  
300 - input verification will be performed concurrently with the  
301 - 1st pass, and you can reject the input according to the  
302 - result as specified by -gmmreject. The GMM should be defined  
303 - as one-state HMMs.  
304 -  
305 - -gmmnum number  
306 - Number of Gaussian components to be computed per frame on GMM  
307 - calculation. Only the N-best Gaussians will be computed for  
308 - rapid calculation. The default is 10 and specifying smaller  
309 - value will speed up GMM calculation, but too small value (1  
310 - or 2) may cause degradation of identification performance.  
311 -  
312 - -gmmreject string  
313 - Comma-separated list of GMM names to be rejected as invalid  
314 - input. When recognition, the log likelihoods of GMMs  
315 - accumulated for the entire input will be computed  
316 - concurrently with the 1st pass. If the GMM name of the  
317 - maximum score is within this string, the 2nd pass will not be  
318 - executed and the input will be rejected.  
319 -  
320 - -gmmmargin frames  
321 - (GMM_VAD) Head margin in frames. When a speech trigger  
322 - detected by GMM, recognition will start from current frame  
323 - minus this value. (Rev.4.0)  
324 -  
325 - This option will be valid only if compiled with  
326 - --enable-gmm-vad.  
327 -  
328 - -gmmup value  
329 - (GMM_VAD) Up trigger threshold of voice activity score.  
330 - (Rev.4.1)  
331 -  
332 - This option will be valid only if compiled with  
333 - --enable-gmm-vad.  
334 -  
335 - -gmmdown value  
336 - (GMM_VAD) Down trigger threshold of voice activity score.  
337 - (Rev.4.1)  
338 -  
339 - This option will be valid only if compiled with  
340 - --enable-gmm-vad.  
341 -  
342 - Decoding option  
343 - Real-time processing means concurrent processing of MFCC  
344 - computation 1st pass decoding. By default, real-time processing  
345 - on the pass is on for microphone / adinnet / netaudio input, and  
346 - for others.  
347 -  
348 - -realtime , -norealtime  
349 - Explicitly switch on / off real-time (pipe-line) processing  
350 - on the first pass. The default is off for file input, and on  
351 - for microphone, adinnet and NetAudio input. This option  
352 - relates to the way CMN and energy normalization is performed:  
353 - if off, they will be done using average features of whole  
354 - input. If on, MAP-CMN and energy normalization to do  
355 - real-time processing.  
356 -  
357 - Misc. options  
358 - -C jconffile  
359 - Load a jconf file at here. The content of the jconffile will  
360 - be expanded at this point.  
361 -  
362 - -version  
363 - Print version information to standard error, and exit.  
364 -  
365 - -setting  
366 - Print engine setting information to standard error, and exit.  
367 -  
368 - -quiet  
369 - Output less log. For result, only the best word sequence will  
370 - be printed.  
371 -  
372 - -debug  
373 - (For debug) output enormous internal message and debug  
374 - information to log.  
375 -  
376 - -check {wchmm|trellis|triphone}  
377 - For debug, enter interactive check mode.  
378 -  
379 - -plugindir dirlist  
380 - Specify directory to load plugin. If several direcotries  
381 - exist, specify them by colon-separated list.  
382 -  
383 - Instance declaration for multi decoding  
384 - The following arguments will create a new configuration set with  
385 - default parameters, and switch current set to it. Jconf parameters  
386 - specified after the option will be set into the current set.  
387 -  
388 - To do multi-model decoding, these argument should be specified at the  
389 - first of each model / search instances with different names. Any  
390 - options before the first instance definition will be IGNORED.  
391 -  
392 - When no instance definition is found (as older version of Julius), all  
393 - the options are assigned to a default instance named _default.  
394 -  
395 - Please note that decoding with a single LM and multiple AMs is not  
396 - fully supported. For example, you may want to construct the jconf file  
397 - as following.  
398 - This type of model sharing is not supported yet, since some part of LM  
399 - processing depends on the assigned AM. Instead, you can get the same  
400 - result by defining the same LMs for each AM, like this:  
401 -  
402 - -AM name  
403 - Create a new AM configuration set, and switch current to the new  
404 - one. You should give a unique name. (Rev.4.0)  
405 -  
406 - -LM name  
407 - Create a new LM configuration set, and switch current to the new  
408 - one. You should give a unique name. (Rev.4.0)  
409 -  
410 - -SR name am_name lm_name  
411 - Create a new search configuration set, and switch current to the new  
412 - one. The specified AM and LM will be assigned to it. The am_name and  
413 - lm_name can be either name or ID number. You should give a unique  
414 - name. (Rev.4.0)  
415 -  
416 - -AM_GMM  
417 - When using GMM for front-end processing, you can specify  
418 - GMM-specific acoustic parameters after this option. If you does not  
419 - specify -AM_GMM with GMM, the GMM will share the same parameter  
420 - vector as the last AM. The current AM will be switched to the GMM  
421 - one, so be careful not to confuse with normal AM configurations.  
422 - (Rev.4.0)  
423 -  
424 - -GLOBAL  
425 - Start a global section. The global options should be placed before  
426 - any instance declaration, or after this option on multiple model  
427 - recognition. This can be used multiple times. (Rev.4.1)  
428 -  
429 - -nosectioncheck , -sectioncheck  
430 - Disable / enable option location check in multi-model decoding. When  
431 - enabled, the options between instance declaration is treated as  
432 - "sections" and only the belonging option types can be written. For  
433 - example, when an option -AM is specified, only the AM related option  
434 - can be placed after the option until other declaration is found.  
435 - Also, global options should be placed at top, before any instance  
436 - declarataion. This is enabled by default. (Rev.4.1)  
437 -  
438 - Language model (-LM)  
439 - This group contains options for model definition of each language model  
440 - type. When using multiple LM, one instance can have only one LM.  
441 -  
442 - Only one type of LM can be specified for a LM configuration. If you  
443 - want to use multi model, you should define them one as a new LM.  
444 -  
445 - N-gram  
446 - -d bingram_file  
447 - Use binary format N-gram. An ARPA N-gram file can be  
448 - converted to Julius binary format by mkbingram.  
449 -  
450 - -nlr arpa_ngram_file  
451 - A forward, left-to-right N-gram language model in standard  
452 - ARPA format. When both a forward N-gram and backward N-gram  
453 - are specified, Julius uses this forward 2-gram for the 1st  
454 - pass, and the backward N-gram for the 2nd pass.  
455 -  
456 - Since ARPA file often gets huge and requires a lot of time to  
457 - load, it may be better to convert the ARPA file to Julius  
458 - binary format by mkbingram. Note that if both forward and  
459 - backward N-gram is used for recognition, they together will  
460 - be converted to a single binary.  
461 -  
462 - When only a forward N-gram is specified by this option and no  
463 - backward N-gram specified by -nrl, Julius performs  
464 - recognition with only the forward N-gram. The 1st pass will  
465 - use the 2-gram entry in the given N-gram, and The 2nd pass  
466 - will use the given N-gram, with converting forward  
467 - probabilities to backward probabilities by Bayes rule.  
468 - (Rev.4.0)  
469 -  
470 - -nrl arpa_ngram_file  
471 - A backward, right-to-left N-gram language model in standard  
472 - ARPA format. When both a forward N-gram and backward N-gram  
473 - are specified, Julius uses the forward 2-gram for the 1st  
474 - pass, and this backward N-gram for the 2nd pass.  
475 -  
476 - Since ARPA file often gets huge and requires a lot of time to  
477 - load, it may be better to convert the ARPA file to Julius  
478 - binary format by mkbingram. Note that if both forward and  
479 - backward N-gram is used for recognition, they together will  
480 - be converted to a single binary.  
481 -  
482 - When only a backward N-gram is specified by this option and  
483 - no forward N-gram specified by -nlr, Julius performs  
484 - recognition with only the backward N-gram. The 1st pass will  
485 - use the forward 2-gram probability computed from the backward  
486 - 2-gram using Bayes rule. The 2nd pass fully use the given  
487 - backward N-gram. (Rev.4.0)  
488 -  
489 - -v dict_file  
490 - Word dictionary file.  
491 -  
492 - -silhead word_string -siltail word_string  
493 - Silence word defined in the dictionary, for silences at the  
494 - beginning of sentence and end of sentence. (default: "<s>",  
495 - "</s>")  
496 -  
497 - -mapunk word_string  
498 - Specify unknown word. Default is "<unk>" or "<UNK>". This  
499 - will be used to assign word probability on unknown words,  
500 - i.e. words in dictionary that are not in N-gram vocabulary.  
501 -  
502 - -iwspword  
503 - Add a word entry to the dictionary that should correspond to  
504 - inter-word pauses. This may improve recognition accuracy in  
505 - some language model that has no explicit inter-word pause  
506 - modeling. The word entry to be added can be changed by  
507 - -iwspentry.  
508 -  
509 - -iwspentry word_entry_string  
510 - Specify the word entry that will be added by -iwspword.  
511 - (default: "<UNK> [sp] sp sp")  
512 -  
513 - -sepnum number  
514 - Number of high frequency words to be isolated from the  
515 - lexicon tree, to ease approximation error that may be caused  
516 - by the one-best approximation on 1st pass. (default: 150)  
517 -  
518 - Grammar  
519 - Multiple grammars can be specified by repeating -gram and  
520 - -gramlist. Note that this is unusual behavior from other options  
521 - (in normal Julius option, last one will override previous ones).  
522 - You can use -nogram to reset the grammars already specified  
523 - before the point.  
524 -  
525 - -gram gramprefix1[,gramprefix2[,gramprefix3,...]]  
526 - Comma-separated list of grammars to be used. the argument  
527 - should be a prefix of a grammar, i.e. if you have foo.dfa and  
528 - foo.dict, you should specify them with a single argument foo.  
529 - Multiple grammars can be specified at a time as a  
530 - comma-separated list.  
531 -  
532 - -gramlist list_file  
533 - Specify a grammar list file that contains list of grammars to  
534 - be used. The list file should contain the prefixes of  
535 - grammars, each per line. A relative path in the list file  
536 - will be treated as relative to the file, not the current path  
537 - or configuration file.  
538 -  
539 - -dfa dfa_file -v dict_file  
540 - An old way of specifying grammar files separately. This is  
541 - bogus, and should not be used any more.  
542 -  
543 - -nogram  
544 - Remove the current list of grammars already specified by  
545 - -gram, -gramlist, -dfa and -v.  
546 -  
547 - Isolated word  
548 - Dictionary can be specified by using -w and -wlist. When you  
549 - specify multiple times, all of them will be read at startup. You  
550 - can use -nogram to reset the already specified dictionaries at  
551 - that point.  
552 -  
553 - -w dict_file  
554 - Word dictionary for isolated word recognition. File format is  
555 - the same as other LM. (Rev.4.0)  
556 -  
557 - -wlist list_file  
558 - Specify a dictionary list file that contains list of  
559 - dictionaries to be used. The list file should contain the  
560 - file name of dictionaries, each per line. A relative path in  
561 - the list file will be treated as relative to the list file,  
562 - not the current path or configuration file. (Rev.4.0)  
563 -  
564 - -nogram  
565 - Remove the current list of dictionaries already specified by  
566 - -w and -wlist.  
567 -  
568 - -wsil head_sil_model_name tail_sil_model_name sil_context_name  
569 - On isolated word recognition, silence models will be appended  
570 - to the head and tail of each word at recognition. This option  
571 - specifies the silence models to be appended.  
572 - sil_context_name is the name of the head sil model and tail  
573 - sil model as a context of word head phone and tail phone. For  
574 - example, if you specify -wsil silB silE sp, a word with phone  
575 - sequence b eh t will be translated as silB sp-b+eh b-eh+t  
576 - eh-t+sp silE. (Rev.4.0)  
577 -  
578 - User-defined LM  
579 - -userlm  
580 - Declare to use user LM functions in the program. This option  
581 - should be specified if you use user-defined LM functions.  
582 - (Rev.4.0)  
583 -  
584 - Misc. LM options  
585 - -forcedict  
586 - Skip error words in dictionary and force running.  
587 -  
588 - Acoustic model and feature analysis (-AM) (-AM_GMM)  
589 - This section is about options for acoustic model, feature extraction,  
590 - feature normalizations and spectral subtraction.  
591 -  
592 - After -AM name, an acoustic model and related specification should be  
593 - written. You can use multiple AMs trained with different MFCC types.  
594 - For GMM, the required parameter condition should be specified just as  
595 - same as AMs after -AM_GMM.  
596 -  
597 - When using multiple AMs, the values of -smpPeriod, -smpFreq, -fsize and  
598 - -fshift should be the same among all AMs.  
599 -  
600 - Acoustic HMM  
601 - -h hmmdef_file  
602 - Acoustic HMM definition file. It should be in HTK ascii  
603 - format, or Julius binary format. You can convert HTK ascii  
604 - format to Julius binary format using mkbinhmm.  
605 -  
606 - -hlist hmmlist_file  
607 - HMMList file for phone mapping. This file provides mapping  
608 - between logical triphone names generated in the dictionary  
609 - and the defined HMM names in hmmdefs. This option should be  
610 - specified for context-dependent model.  
611 -  
612 - -tmix number  
613 - Specify the number of top Gaussians to be calculated in a  
614 - mixture codebook. Small number will speed up the acoustic  
615 - computation, but AM accuracy may get worse with too small  
616 - value. See also -gprune. (default: 2)  
617 -  
618 - -spmodel name  
619 - Specify HMM model name that corresponds to short-pause in an  
620 - utterance. The short-pause model name will be used in  
621 - recognition: short-pause skipping on grammar recognition,  
622 - word-end short-pause model insertion with -iwsp on N-gram, or  
623 - short-pause segmentation (-spsegment). (default: "sp")  
624 -  
625 - -multipath  
626 - Enable multi-path mode. To make decoding faster, Julius by  
627 - default impose a limit on HMM transitions that each model  
628 - should have only one transition from initial state and to end  
629 - state. On multi-path mode, Julius does extra handling on  
630 - inter-model transition to allows model-skipping transition  
631 - and multiple output/input transitions. Note that specifying  
632 - this option will make Julius a bit slower, and the larger  
633 - beam width may be required.  
634 -  
635 - This function was a compilation-time option on Julius 3.x,  
636 - and now becomes a run-time option. By default (without this  
637 - option), Julius checks the transition type of specified HMMs,  
638 - and enable the multi-path mode if required. You can force  
639 - multi-path mode with this option. (rev.4.0)  
640 -  
641 - -gprune {safe|heuristic|beam|none|default}  
642 - Set Gaussian pruning algorithm to use. For tied-mixture  
643 - model, Julius performs Gaussian pruning to reduce acoustic  
644 - computation, by calculating only the top N Gaussians in each  
645 - codebook at each frame. The default setting will be set  
646 - according to the model type and engine setting. default will  
647 - force accepting the default setting. Set this to none to  
648 - disable pruning and perform full computation. safe  
649 - guarantees the top N Gaussians to be computed. heuristic and  
650 - beam do more aggressive computational cost reduction, but may  
651 - result in small loss of accuracy model (default: safe  
652 - (standard), beam (fast) for tied mixture model, none for non  
653 - tied-mixture model).  
654 -  
655 - -iwcd1 {max|avg|best number}  
656 - Select method to approximate inter-word triphone on the head  
657 - and tail of a word in the first pass.  
658 -  
659 -  
660 - max will apply the maximum likelihood of the same context  
661 - triphones. avg will apply the average likelihood of the same  
662 - context triphones. best number will apply the average of top  
663 - N-best likelihoods of the same context triphone.  
664 -  
665 - Default is best 3 for use with N-gram, and avg for grammar  
666 - and word. When this AM is shared by LMs of both type, latter  
667 - one will be chosen.  
668 -  
669 - -iwsppenalty float  
670 - Insertion penalty for word-end short pauses appended by  
671 - -iwsp.  
672 -  
673 - -gshmm hmmdef_file  
674 - If this option is specified, Julius performs Gaussian Mixture  
675 - Selection for efficient decoding. The hmmdefs should be a  
676 - monophone model generated from an ordinary monophone HMM  
677 - model, using mkgshmm.  
678 -  
679 - -gsnum number  
680 - On GMS, specify number of monophone states to compute  
681 - corresponding triphones in detail. (default: 24)  
682 -  
683 - Speech analysis  
684 - Only MFCC feature extraction is supported in current Julius.  
685 - Thus when recognizing a waveform input from file or microphone,  
686 - AM must be trained by MFCC. The parameter condition should also  
687 - be set as exactly the same as the training condition by the  
688 - options below.  
689 -  
690 - When you give an input in HTK Parameter file, you can use any  
691 - parameter type for AM. In this case Julius does not care about  
692 - the type of input feature and AM, just read them as vector  
693 - sequence and match them to the given AM. Julius only checks  
694 - whether the parameter types are the same. If it does not work  
695 - well, you can disable this checking by -notypecheck.  
696 -  
697 - In Julius, the parameter kind and qualifiers (as TARGETKIND in  
698 - HTK) and the number of cepstral parameters (NUMCEPS) will be set  
699 - automatically from the content of the AM header, so you need not  
700 - specify them by options.  
701 -  
702 - Other parameters should be set exactly the same as training  
703 - condition. You can also give a HTK Config file which you used to  
704 - train AM to Julius by -htkconf. When this option is applied,  
705 - Julius will parse the Config file and set appropriate parameter.  
706 -  
707 - You can further embed those analysis parameter settings to a  
708 - binary HMM file using mkbinhmm.  
709 -  
710 - If options specified in several ways, they will be evaluated in  
711 - the order below. The AM embedded parameter will be loaded first  
712 - if any. Then, the HTK config file given by -htkconf will be  
713 - parsed. If a value already set by AM embedded value, HTK config  
714 - will override them. At last, the direct options will be loaded,  
715 - which will override settings loaded before. Note that, when the  
716 - same options are specified several times, later will override  
717 - previous, except that -htkconf will be evaluated first as  
718 - described above.  
719 -  
720 - -smpPeriod period  
721 - Sampling period of input speech, in unit of 100 nanoseconds.  
722 - Sampling rate can also be specified by -smpFreq. Please note  
723 - that the input frequency should be set equal to the training  
724 - conditions of AM. (default: 625, corresponds to 16,000Hz)  
725 -  
726 - This option corresponds to the HTK Option SOURCERATE. The  
727 - same value can be given to this option.  
728 -  
729 - When using multiple AM, this value should be the same among  
730 - all AMs.  
731 -  
732 - -smpFreq Hz  
733 - Set sampling frequency of input speech in Hz. Sampling rate  
734 - can also be specified using -smpPeriod. Please note that this  
735 - frequency should be set equal to the training conditions of  
736 - AM. (default: 16,000)  
737 -  
738 - When using multiple AM, this value should be the same among  
739 - all AMs.  
740 -  
741 - -fsize sample_num  
742 - Window size in number of samples. (default: 400)  
743 -  
744 - This option corresponds to the HTK Option WINDOWSIZE, but  
745 - value should be in samples (HTK value / smpPeriod).  
746 -  
747 - When using multiple AM, this value should be the same among  
748 - all AMs.  
749 -  
750 - -fshift sample_num  
751 - Frame shift in number of samples. (default: 160)  
752 -  
753 - This option corresponds to the HTK Option TARGETRATE, but  
754 - value should be in samples (HTK value / smpPeriod).  
755 -  
756 - When using multiple AM, this value should be the same among  
757 - all AMs.  
758 -  
759 - -preemph float  
760 - Pre-emphasis coefficient. (default: 0.97)  
761 -  
762 - This option corresponds to the HTK Option PREEMCOEF. The same  
763 - value can be given to this option.  
764 -  
765 - -fbank num  
766 - Number of filterbank channels. (default: 24)  
767 -  
768 - This option corresponds to the HTK Option NUMCHANS. The same  
769 - value can be given to this option. Be aware that the default  
770 - value not the same as in HTK (22).  
771 -  
772 - -ceplif num  
773 - Cepstral liftering coefficient. (default: 22)  
774 -  
775 - This option corresponds to the HTK Option CEPLIFTER. The same  
776 - value can be given to this option.  
777 -  
778 - -rawe , -norawe  
779 - Enable/disable using raw energy before pre-emphasis (default:  
780 - disabled)  
781 -  
782 - This option corresponds to the HTK Option RAWENERGY. Be aware  
783 - that the default value differs from HTK (enabled at HTK,  
784 - disabled at Julius).  
785 -  
786 - -enormal , -noenormal  
787 - Enable/disable normalizing log energy. On live input, this  
788 - normalization will be approximated from the average of last  
789 - input. (default: disabled)  
790 -  
791 - This option corresponds to the HTK Option ENORMALISE. Be  
792 - aware that the default value differs from HTK (enabled at  
793 - HTK, disabled at Julius).  
794 -  
795 - -escale float_scale  
796 - Scaling factor of log energy when normalizing log energy.  
797 - (default: 1.0)  
798 -  
799 - This option corresponds to the HTK Option ESCALE. Be aware  
800 - that the default value differs from HTK (0.1).  
801 -  
802 - -silfloor float  
803 - Energy silence floor in dB when normalizing log energy.  
804 - (default: 50.0)  
805 -  
806 - This option corresponds to the HTK Option SILFLOOR.  
807 -  
808 - -delwin frame  
809 - Delta window size in number of frames. (default: 2)  
810 -  
811 - This option corresponds to the HTK Option DELTAWINDOW. The  
812 - same value can be given to this option.  
813 -  
814 - -accwin frame  
815 - Acceleration window size in number of frames. (default: 2)  
816 -  
817 - This option corresponds to the HTK Option ACCWINDOW. The same  
818 - value can be given to this option.  
819 -  
820 - -hifreq Hz  
821 - Enable band-limiting for MFCC filterbank computation: set  
822 - upper frequency cut-off. Value of -1 will disable it.  
823 - (default: -1)  
824 -  
825 - This option corresponds to the HTK Option HIFREQ. The same  
826 - value can be given to this option.  
827 -  
828 - -lofreq Hz  
829 - Enable band-limiting for MFCC filterbank computation: set  
830 - lower frequency cut-off. Value of -1 will disable it.  
831 - (default: -1)  
832 -  
833 - This option corresponds to the HTK Option LOFREQ. The same  
834 - value can be given to this option.  
835 -  
836 - -zmeanframe , -nozmeanframe  
837 - With speech input, this option enables/disables frame-wise DC  
838 - offset removal. This corresponds to HTK configuration  
839 - ZMEANSOURCE. This cannot be used together with -zmean.  
840 - (default: disabled)  
841 -  
842 - -usepower  
843 - Use power instead of magnitude on filterbank analysis.  
844 - (default: disabled)  
845 -  
846 - Normalization  
847 - Julius can perform cepstral mean normalization (CMN) for inputs.  
848 - CMN will be activated when the given AM was trained with CMN  
849 - (i.e. has "_Z" qualifier in the header).  
850 -  
851 - The cepstral mean will be estimated in different way according  
852 - to the input type. On file input, the mean will be computed from  
853 - the whole input. On live input such as microphone and network  
854 - input, the ceptral mean of the input is unknown at the start. So  
855 - MAP-CMN will be used. On MAP-CMN, an initial mean vector will be  
856 - applied at the beginning, and the mean vector will be smeared to  
857 - the mean of the incrementing input vector as input goes. Options  
858 - below can control the behavior of MAP-CMN.  
859 -  
860 - -cvn  
861 - Enable cepstral variance normalization. At file input, the  
862 - variance of whole input will be calculated and then applied.  
863 - At live microphone input, variance of the last input will be  
864 - applied. CVN is only supported for an audio input.  
865 -  
866 - -vtln alpha lowcut hicut  
867 - Do frequency warping, typically for a vocal tract length  
868 - normalization (VTLN). Arguments are warping factor, high  
869 - frequency cut-off and low freq. cut-off. They correspond to  
870 - HTK Config values, WARPFREQ, WARPHCUTOFF and WARPLCUTOFF.  
871 -  
872 - -cmnload file  
873 - Load initial cepstral mean vector from file on startup. The  
874 - file should be one saved by -cmnsave. Loading an initial  
875 - cepstral mean enables Julius to better recognize the first  
876 - utterance on a real-time input. When used together with  
877 - -cmnnoupdate, this initial value will be used for all input.  
878 -  
879 - -cmnsave file  
880 - Save the calculated cepstral mean vector into file. The  
881 - parameters will be saved at each input end. If the output  
882 - file already exists, it will be overridden.  
883 -  
884 - -cmnupdate -cmnnoupdate  
885 - Control whether to update the cepstral mean at each input on  
886 - real-time input. Disabling this and specifying -cmnload will  
887 - make engine to always use the loaded static initial cepstral  
888 - mean.  
889 -  
890 - -cmnmapweight float  
891 - Specify the weight of initial cepstral mean for MAP-CMN.  
892 - Specify larger value to retain the initial cepstral mean for  
893 - a longer period, and smaller value to make the cepstral mean  
894 - rely more on the current input. (default: 100.0)  
895 -  
896 - Front-end processing  
897 - Julius can perform spectral subtraction to reduce some  
898 - stationary noise from audio input. Though it is not a powerful  
899 - method, but it may work on some situation. Julius has two ways  
900 - to estimate noise spectrum. One way is to assume that the first  
901 - short segment of an speech input is noise segment, and estimate  
902 - the noise spectrum as the average of the segment. Another way is  
903 - to calculate average spectrum from noise-only input using other  
904 - tool mkss, and load it in Julius. The former one is popular for  
905 - speech file input, and latter should be used in live input. The  
906 - options below will switch / control the behavior.  
907 -  
908 - -sscalc  
909 - Perform spectral subtraction using head part of each file as  
910 - silence part. The head part length should be specified by  
911 - -sscalclen. Valid only for file input. Conflict with -ssload.  
912 -  
913 - -sscalclen msec  
914 - With -sscalc, specify the length of head silence for noise  
915 - spectrum estimation in milliseconds. (default: 300)  
916 -  
917 - -ssload file  
918 - Perform spectral subtraction for speech input using  
919 - pre-estimated noise spectrum loaded from file. The noise  
920 - spectrum file can be made by mkss. Valid for all speech  
921 - input. Conflict with -sscalc.  
922 -  
923 - -ssalpha float  
924 - Alpha coefficient of spectral subtraction for -sscalc and  
925 - -ssload. Noise will be subtracted stronger as this value gets  
926 - larger, but distortion of the resulting signal also becomes  
927 - remarkable. (default: 2.0)  
928 -  
929 - -ssfloor float  
930 - Flooring coefficient of spectral subtraction. The spectral  
931 - power that goes below zero after subtraction will be  
932 - substituted by the source signal with this coefficient  
933 - multiplied. (default: 0.5)  
934 -  
935 - Misc. AM options  
936 - -htkconf file  
937 - Parse the given HTK Config file, and set corresponding  
938 - parameters to Julius. When using this option, the default  
939 - parameter values are switched from Julius defaults to HTK  
940 - defaults.  
941 -  
942 - Recognition process and search (-SR)  
943 - This section contains options for search parameters on the 1st / 2nd  
944 - pass such as beam width and LM weights, configurations for short-pause  
945 - segmentation, switches for word lattice output and confusion network  
946 - output, forced alignments, and other options relating recognition  
947 - process and result output.  
948 -  
949 - Default values for beam width and LM weights will change according to  
950 - compile-time setup of JuliusLib , AM model type, and LM size. Please  
951 - see the startup log for the actual values.  
952 -  
953 - 1st pass parameters  
954 - -lmp weight penalty  
955 - (N-gram) Language model weights and word insertion penalties  
956 - for the first pass.  
957 -  
958 - -penalty1 penalty  
959 - (Grammar) word insertion penalty for the first pass.  
960 - (default: 0.0)  
961 -  
962 - -b width  
963 - Beam width in number of HMM nodes for rank beaming on the  
964 - first pass. This value defines search width on the 1st pass,  
965 - and has dominant effect on the total processing time. Smaller  
966 - width will speed up the decoding, but too small value will  
967 - result in a substantial increase of recognition errors due to  
968 - search failure. Larger value will make the search stable and  
969 - will lead to failure-free search, but processing time will  
970 - grow in proportion to the width.  
971 -  
972 - The default value is dependent on acoustic model type: 400  
973 - (monophone), 800 (triphone), or 1000 (triphone, setup=v2.1)  
974 -  
975 - -nlimit num  
976 - Upper limit of token per node. This option is valid when  
977 - --enable-wpair and --enable-wpair-nlimit are enabled at  
978 - compilation time.  
979 -  
980 - -progout  
981 - Enable progressive output of the partial results on the first  
982 - pass.  
983 -  
984 - -proginterval msec  
985 - Set the time interval for -progout in milliseconds. (default:  
986 - 300)  
987 -  
988 - 2nd pass parameters  
989 - -lmp2 weight penalty  
990 - (N-gram) Language model weights and word insertion penalties  
991 - for the second pass.  
992 -  
993 - -penalty2 penalty  
994 - (Grammar) word insertion penalty for the second pass.  
995 - (default: 0.0)  
996 -  
997 - -b2 width  
998 - Envelope beam width (number of hypothesis) at the second  
999 - pass. If the count of word expansion at a certain hypothesis  
1000 - length reaches this limit while search, shorter hypotheses  
1001 - are not expanded further. This prevents search to fall in  
1002 - breadth-first-like situation stacking on the same position,  
1003 - and improve search failure mostly for large vocabulary  
1004 - condition. (default: 30)  
1005 -  
1006 - -sb float  
1007 - Score envelope width for enveloped scoring. When calculating  
1008 - hypothesis score for each generated hypothesis, its trellis  
1009 - expansion and Viterbi operation will be pruned in the middle  
1010 - of the speech if score on a frame goes under the width.  
1011 - Giving small value makes the second pass faster, but  
1012 - computation error may occur. (default: 80.0)  
1013 -  
1014 - -s num  
1015 - Stack size, i.e. the maximum number of hypothesis that can be  
1016 - stored on the stack during the search. A larger value may  
1017 - give more stable results, but increases the amount of memory  
1018 - required. (default: 500)  
1019 -  
1020 - -m count  
1021 - Number of expanded hypotheses required to discontinue the  
1022 - search. If the number of expanded hypotheses is greater then  
1023 - this threshold then, the search is discontinued at that  
1024 - point. The larger this value is, The longer Julius gets to  
1025 - give up search. (default: 2000)  
1026 -  
1027 - -n num  
1028 - The number of candidates Julius tries to find. The search  
1029 - continues till this number of sentence hypotheses have been  
1030 - found. The obtained sentence hypotheses are sorted by score,  
1031 - and final result is displayed in the order (see also the  
1032 - -output). The possibility that the optimum hypothesis is  
1033 - correctly found increases as this value gets increased, but  
1034 - the processing time also becomes longer. The default value  
1035 - depends on the engine setup on compilation time: 10  
1036 - (standard) or 1 (fast or v2.1)  
1037 -  
1038 - -output num  
1039 - The top N sentence hypothesis to be output at the end of  
1040 - search. Use with -n (default: 1)  
1041 -  
1042 - -lookuprange frame  
1043 - Set the number of frames before and after to look up next  
1044 - word hypotheses in the word trellis on the second pass. This  
1045 - prevents the omission of short words, but with a large value,  
1046 - the number of expanded hypotheses increases and system  
1047 - becomes slow. (default: 5)  
1048 -  
1049 - -looktrellis  
1050 - (Grammar) Expand only the words survived on the first pass  
1051 - instead of expanding all the words predicted by grammar. This  
1052 - option makes second pass decoding faster especially for large  
1053 - vocabulary condition, but may increase deletion error of  
1054 - short words. (default: disabled)  
1055 -  
1056 - Short-pause segmentation / decoder-VAD  
1057 - When compiled with --enable-decoder-vad, the short-pause  
1058 - segmentation will be extended to support decoder-based VAD.  
1059 -  
1060 - -spsegment  
1061 - Enable short-pause segmentation mode. Input will be segmented  
1062 - when a short pause word (word with only silence model in  
1063 - pronunciation) gets the highest likelihood at certain  
1064 - successive frames on the first pass. When detected segment  
1065 - end, Julius stop the 1st pass at the point, perform 2nd pass,  
1066 - and continue with next segment. The word context will be  
1067 - considered among segments. (Rev.4.0)  
1068 -  
1069 - When compiled with --enable-decoder-vad, this option enables  
1070 - decoder-based VAD, to skip long silence.  
1071 -  
1072 - -spdur frame  
1073 - Short pause duration length to detect end of input segment,  
1074 - in number of frames. (default: 10)  
1075 -  
1076 - -pausemodels string  
1077 - A comma-separated list of pause model names to be used at  
1078 - short-pause segmentation. The word whose pronunciation  
1079 - consists of only the pause models will be treated as "pause  
1080 - word" and used for pause detection. If not specified, name of  
1081 - -spmodel, -silhead and -siltail will be used. (Rev.4.0)  
1082 -  
1083 - -spmargin frame  
1084 - Back step margin at trigger up for decoder-based VAD. When  
1085 - speech up-trigger found by decoder-VAD, Julius will rewind  
1086 - the input parameter by this value, and start recognition at  
1087 - the point. (Rev.4.0)  
1088 -  
1089 - This option will be valid only if compiled with  
1090 - --enable-decoder-vad.  
1091 -  
1092 - -spdelay frame  
1093 - Trigger decision delay frame at trigger up for decoder-based  
1094 - VAD. (Rev.4.0)  
1095 -  
1096 - This option will be valid only if compiled with  
1097 - --enable-decoder-vad.  
1098 -  
1099 - Word lattice / confusion network output  
1100 - -lattice , -nolattice  
1101 - Enable / disable generation of word graph. Search algorithm  
1102 - also has changed to optimize for better word graph  
1103 - generation, so the sentence result may not be the same as  
1104 - normal N-best recognition. (Rev.4.0)  
1105 -  
1106 - -confnet , -noconfnet  
1107 - Enable / disable generation of confusion network. Enabling  
1108 - this will also activates -lattice internally. (Rev.4.0)  
1109 -  
1110 - -graphrange frame  
1111 - Merge same words at neighbor position at graph generation. If  
1112 - the beginning time and ending time of two word candidates of  
1113 - the same word is within the specified range, they will be  
1114 - merged. The default is 0 (allow merging same words on exactly  
1115 - the same location) and specifying larger value will result in  
1116 - smaller graph output. Setting this value to -1 will disable  
1117 - merging, in that case same words on the same location of  
1118 - different scores will be left as they are. (default: 0)  
1119 -  
1120 - -graphcut depth  
1121 - Cut the resulting graph by its word depth at post-processing  
1122 - stage. The depth value is the number of words to be allowed  
1123 - at a frame. Setting to -1 disables this feature. (default:  
1124 - 80)  
1125 -  
1126 - -graphboundloop count  
1127 - Limit the number of boundary adjustment loop at  
1128 - post-processing stage. This parameter prevents Julius from  
1129 - blocking by infinite adjustment loop by short word  
1130 - oscillation. (default: 20)  
1131 -  
1132 - -graphsearchdelay , -nographsearchdelay  
1133 - When this option is enabled, Julius modifies its graph  
1134 - generation algorithm on the 2nd pass not to terminate search  
1135 - by graph merging, until the first sentence candidate is  
1136 - found. This option may improve graph accuracy, especially  
1137 - when you are going to generate a huge word graph by setting  
1138 - broad search. Namely, it may result in better graph accuracy  
1139 - when you set wide beams on both 1st pass -b and 2nd pass -b2,  
1140 - and large number for -n. (default: disabled)  
1141 -  
1142 - Multi-gram / multi-dic recognition  
1143 - -multigramout , -nomultigramout  
1144 - On grammar recognition using multiple grammars, Julius will  
1145 - output only the best result among all grammars. Enabling this  
1146 - option will make Julius to output result for each grammar.  
1147 - (default: disabled)  
1148 -  
1149 - Forced alignment  
1150 - -walign  
1151 - Do viterbi alignment per word units for the recognition  
1152 - result. The word boundary frames and the average acoustic  
1153 - scores per frame will be calculated.  
1154 -  
1155 - -palign  
1156 - Do viterbi alignment per phone units for the recognition  
1157 - result. The phone boundary frames and the average acoustic  
1158 - scores per frame will be calculated.  
1159 -  
1160 - -salign  
1161 - Do viterbi alignment per state for the recognition result.  
1162 - The state boundary frames and the average acoustic scores per  
1163 - frame will be calculated.  
1164 -  
1165 - Misc. search options  
1166 - -inactive  
1167 - Start this recognition process instance with inactive state.  
1168 - (Rev.4.0)  
1169 -  
1170 - -1pass  
1171 - Perform only the first pass.  
1172 -  
1173 - -fallback1pass  
1174 - When 2nd pass fails, Julius finish the recognition with no  
1175 - result. This option tell Julius to output the 1st pass result  
1176 - as a final result when the 2nd pass fails. Note that some  
1177 - score output (confidence etc.) may not be useful. This was  
1178 - the default behavior of Julius-3.x.  
1179 -  
1180 - -no_ccd , -force_ccd  
1181 - Explicitly switch phone context handling at search. Normally  
1182 - Julius determines whether the using AM is a context-dependent  
1183 - model or not from the model names, i.e., whether the names  
1184 - contain character + and -. This option will override the  
1185 - automatic detection.  
1186 -  
1187 - -cmalpha float  
1188 - Smoothing parameter for confidence scoring. (default: 0.05)  
1189 -  
1190 - -iwsp  
1191 - (Multi-path mode only) Enable inter-word context-free short  
1192 - pause insertion. This option appends a skippable short pause  
1193 - model for every word end. The short-pause model can be  
1194 - specified by -spmodel.  
1195 -  
1196 - -transp float  
1197 - Additional insertion penalty for transparent words. (default:  
1198 - 0.0)  
1199 -  
1200 - -demo  
1201 - Equivalent to -progout -quiet.  
1202 -  
1203 -ENVIRONMENT VARIABLES  
1204 - ALSADEV  
1205 - (using mic input with alsa device) specify a capture device name. If  
1206 - not specified, "default" will be used.  
1207 -  
1208 - AUDIODEV  
1209 - (using mic input with oss device) specify a capture device path. If  
1210 - not specified, "/dev/dsp" will be used.  
1211 -  
1212 - LATENCY_MSEC  
1213 - Try to set input latency of microphone input in milliseconds.  
1214 - Smaller value will shorten latency but sometimes make process  
1215 - unstable. Default value will depend on the running OS.  
1216 -  
1217 -EXAMPLES  
1218 - For examples of system usage, refer to the tutorial section in the  
1219 - Julius documents.  
1220 -  
1221 -NOTICE  
1222 - Note about jconf files: relative paths in a jconf file are interpreted  
1223 - as relative to the jconf file itself, not to the current directory.  
1224 -  
1225 -SEE ALSO  
1226 - julian(1), jcontrol(1), adinrec(1), adintool(1), mkbingram(1),  
1227 - mkbinhmm(1), mkgsmm(1), wav2mfcc(1), mkss(1)  
1228 -  
1229 - http://julius.sourceforge.jp/en/  
1230 -  
1231 -DIAGNOSTICS  
1232 - Julius normally will return the exit status 0. If an error occurs,  
1233 - Julius exits abnormally with exit status 1. If an input file cannot be  
1234 - found or cannot be loaded for some reason then Julius will skip  
1235 - processing for that file.  
1236 -  
1237 -BUGS  
1238 - There are some restrictions to the type and size of the models Julius  
1239 - can use. For a detailed explanation refer to the Julius documentation.  
1240 - For bug-reports, inquires and comments please contact julius-info at  
1241 - lists.sourceforge.jp.  
1242 -  
1243 -COPYRIGHT  
1244 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
1245 -  
1246 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
1247 -  
1248 - Copyright (c) 2000-2008 Shikano Lab., Nara Institute of Science and  
1249 - Technology  
1250 -  
1251 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
1252 - Technology  
1253 -  
1254 -AUTHORS  
1255 - Rev.1.0 (1998/02/20)  
1256 - Designed by Tatsuya KAWAHARA and Akinobu LEE (Kyoto University)  
1257 -  
1258 - Development by Akinobu LEE (Kyoto University)  
1259 -  
1260 - Rev.1.1 (1998/04/14), Rev.1.2 (1998/10/31), Rev.2.0 (1999/02/20),  
1261 - Rev.2.1 (1999/04/20), Rev.2.2 (1999/10/04), Rev.3.0 (2000/02/14),  
1262 - Rev.3.1 (2000/05/11)  
1263 - Development of above versions by Akinobu LEE (Kyoto University)  
1264 -  
1265 - Rev.3.2 (2001/08/15), Rev.3.3 (2002/09/11), Rev.3.4 (2003/10/01),  
1266 - Rev.3.4.1 (2004/02/25), Rev.3.4.2 (2004/04/30)  
1267 - Development of above versions by Akinobu LEE (Nara Institute of  
1268 - Science and Technology)  
1269 -  
1270 - Rev.3.5 (2005/11/11), Rev.3.5.1 (2006/03/31), Rev.3.5.2 (2006/07/31),  
1271 - Rev.3.5.3 (2006/12/29), Rev.4.0 (2007/12/19), Rev.4.1 (2008/10/03)  
1272 - Development of above versions by Akinobu LEE (Nagoya Institute of  
1273 - Technology)  
1274 -  
1275 -THANKS TO  
1276 - From rev.3.2, Julius is released by the "Information Processing  
1277 - Society, Continuous Speech Consortium".  
1278 -  
1279 - The Windows DLL version was developed and released by Hideki BANNO  
1280 - (Nagoya University).  
1281 -  
1282 - The Windows Microsoft Speech API compatible version was developed by  
1283 - Takashi SUMIYOSHI (Kyoto University).  
1284 -  
1285 -  
1286 -  
1287 - 02/11/2009 JULIUS(1)  
recognize/src/julius/doc/manuals/mkbingram.txt
View file @e1cf2b2
@@ -1,97 +0,0 @@ @@ -1,97 +0,0 @@
1 - mkbingram  
2 -  
3 -MKBINGRAM(1) MKBINGRAM(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - mkbingram  
9 - - make binary N-gram from ARPA N-gram file  
10 -  
11 -SYNOPSIS  
12 - mkbingram [-nlr forward_ngram.arpa] [-nrl backward_ngram.arpa]  
13 - [-d old_bingram_file] {output_bingram_file}  
14 -  
15 -DESCRIPTION  
16 - mkbingram is a tool to convert N-gram definition file(s) in ARPA  
17 - standard format to a compact Julius binary format. It will speed up the  
18 - initial loading time of N-gram much faster. It can read gzipped file  
19 - directly.  
20 -  
21 - From rev.4.0, Julius can deal with forward N-gram, backward N-gram and  
22 - their combinations. So, mkbingram now generates binary N-gram file from  
23 - one of them, or combining them two to produce one binary N-gram.  
24 -  
25 - When only a forward N-gram is specified, mkbingram generates binary  
26 - N-gram from only the forward N-gram. When using this binary N-gram at  
27 - Julius, it performs the 1st pass with the 2-gram probabilities in the  
28 - N-gram, and run the 2nd pass with the given N-gram fully, with  
29 - converting forward probabilities to backward probabilities by Bayes  
30 - rule.  
31 -  
32 - When only a backward N-gram is specified, mkbingram generates an binary  
33 - N-gram file that contains only the backward N-gram. The 1st pass will  
34 - use forward 2-gram probabilities that can be computed from the backward  
35 - 2-gram using Bayes rule, and the 2nd pass use the given backward N-gram  
36 - fully.  
37 -  
38 - When both forward and backward N-grams are specified, the 2-gram part  
39 - in the forward N-gram and all backward N-gram will be combined into  
40 - single bingram file. The forward 2-gram will be applied for the 1st  
41 - pass and backward N-gram for the 2nd pass. Note that both N-gram should  
42 - be trained in the same corpus with same parameters (i.e. cut-off  
43 - thresholds), with same vocabulary.  
44 -  
45 - The old binary N-gram produced by mkbingram of version 3.x and earlier  
46 - can be used in Julius-4, but you can convert the old version to the new  
47 - version by specifying it as input of current mkbingram by option "-d".  
48 -  
49 - Please note that binary N-gram file converted by mkbingram of version  
50 - 4.0 and later cannot be read by older Julius 3.x.  
51 -  
52 -OPTIONS  
53 - -nlr forward_ngram.arpa  
54 - Read in a forward (left-to-right) word N-gram file in ARPA standard  
55 - format.  
56 -  
57 - -nrl backward_ngram.arpa  
58 - Read in a backward (right-to-left) word N-gram file in ARPA standard  
59 - format.  
60 -  
61 - -d old_bingram_file  
62 - Read in a binary N-gram file.  
63 -  
64 - -swap  
65 - Swap BOS word <s> and EOS word </s> in N-gram.  
66 -  
67 - output_bingram_file  
68 - binary N-gram file name to output.  
69 -  
70 -EXAMPLES  
71 - Convert a set of forward and backward N-gram in ARPA format into Julius  
72 - binary form:  
73 - Convert a single forward 4-gram in ARPA format into a binary file:  
74 - Convert old binary N-gram file to current format:  
75 -  
76 -SEE ALSO  
77 - julius ( 1 ) ,  
78 - mkbinhmm ( 1 ) ,  
79 - mkbinhmmlist ( 1 )  
80 -  
81 -COPYRIGHT  
82 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
83 -  
84 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
85 -  
86 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
87 - Technology  
88 -  
89 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
90 - Technology  
91 -  
92 -LICENSE  
93 - The same as Julius.  
94 -  
95 -  
96 -  
97 - 02/11/2009 MKBINGRAM(1)  
recognize/src/julius/doc/manuals/mkbinhmm.txt
View file @e1cf2b2
@@ -1,78 +0,0 @@ @@ -1,78 +0,0 @@
1 - mkbinhmm  
2 -  
3 -MKBINHMM(1) MKBINHMM(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - mkbinhmm  
9 - - convert HMM definition file in HTK ascii format to Julius binary  
10 - format  
11 -  
12 -SYNOPSIS  
13 - mkbinhmm [-htkconf HTKConfigFile] {hmmdefs_file} {binhmm_file}  
14 -  
15 -DESCRIPTION  
16 - mkbinhmm convert an HMM definition file in HTK ascii format into a  
17 - binary HMM file for Julius. It will greatly speed up the launch  
18 - process.  
19 -  
20 - You can also embed acoustic analysis condition parameters needed for  
21 - recognition into the output file. To embed the parameters, specify the  
22 - HTK Config file you have used to extract acoustic features for training  
23 - the HMM by the optione "-htkconf".  
24 -  
25 - The embedded parameters in a binary HMM format will be loaded into  
26 - Julius automatically, so you do not need to specify the acoustic  
27 - feature options at run time. It will be convenient when you deliver an  
28 - acoustic model.  
29 -  
30 - You can also specify binary file as the input. This can be used to  
31 - update the old binary format into new one, or to embed the config  
32 - parameters into the already existing binary files. If the input binhmm  
33 - already has acoustic analysis parameters embedded, they will be  
34 - overridden by the specified values.  
35 -  
36 -  
37 - mkbinhmm can read gzipped file as input.  
38 -  
39 -OPTIONS  
40 - -htkconf HTKConfigFile  
41 - HTK Config file you used at training time. If specified, the values  
42 - are embedded to the output file.  
43 -  
44 - hmmdefs_file  
45 - The source HMm definitino file in HTK ascii format or Julius binary  
46 - format.  
47 -  
48 - hmmdefs_file  
49 - Output file.  
50 -  
51 -EXAMPLES  
52 - Convert HTK ascii format HMM definition file into Julius binary file:  
53 - Furthermore, embed acoustic feature parameters as specified by Config  
54 - file  
55 - Embed the acoustic parameters into an existing binary file  
56 -  
57 -SEE ALSO  
58 - julius ( 1 ) ,  
59 - mkbingram ( 1 ) ,  
60 - mkbinhmmlist ( 1 )  
61 -  
62 -COPYRIGHT  
63 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
64 -  
65 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
66 -  
67 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
68 - Technology  
69 -  
70 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
71 - Technology  
72 -  
73 -LICENSE  
74 - The same as Julius.  
75 -  
76 -  
77 -  
78 - 10/02/2008 MKBINHMM(1)  
recognize/src/julius/doc/manuals/mkbinhmmlist.txt
View file @e1cf2b2
@@ -1,64 +0,0 @@ @@ -1,64 +0,0 @@
1 - mkbinhmmlist  
2 -  
3 -MKBINHMMLIST(1) MKBINHMMLIST(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - mkbinhmmlist  
9 - - convert HMMList file into binary format  
10 -  
11 -SYNOPSIS  
12 - mkbinhmmlist {hmmdefs_file} {HMMList_file} {output_binhmmlist_file}  
13 -  
14 -DESCRIPTION  
15 - mkbinhmmlist converts a HMMList file to binary format. Since the index  
16 - trees for lookup are also stored in the binary format, it will speed up  
17 - the startup of Julius, namely when using big HMMList file.  
18 -  
19 - For conversion, HMM definition file hmmdefs_file that will be used  
20 - together at Julius needs to be specified. The format of the HMM  
21 - definition file can be either ascii or Julius binary format.  
22 -  
23 - The output binary file can be used in Julius as the same by "-hlist".  
24 - The format wil be auto-detected by Julius.  
25 -  
26 -  
27 - mkbinhmmlist can read gzipped file.  
28 -  
29 -OPTIONS  
30 - hmmdefs_file  
31 - Acoustic HMM definition file, in HMM ascii format or Julius binary  
32 - format.  
33 -  
34 - HMMList_file  
35 - Source HMMList file  
36 -  
37 - output_binhmmlist_file  
38 - Output file, will be overwritten if already exist.  
39 -  
40 -EXAMPLES  
41 - Convert a HMMList file logicalTri into binary format and store to  
42 - logicalTri.bin:  
43 -  
44 -SEE ALSO  
45 - julius ( 1 ) ,  
46 - mkbinhmm ( 1 )  
47 -  
48 -COPYRIGHT  
49 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
50 -  
51 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
52 -  
53 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
54 - Technology  
55 -  
56 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
57 - Technology  
58 -  
59 -LICENSE  
60 - The same as Julius.  
61 -  
62 -  
63 -  
64 - 10/02/2008 MKBINHMMLIST(1)  
recognize/src/julius/doc/manuals/mkdfa.pl.txt
View file @e1cf2b2
@@ -1,67 +0,0 @@ @@ -1,67 +0,0 @@
1 - mkdfa.pl  
2 -  
3 -MKDFA.PL(1) MKDFA.PL(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - mkdfa.pl  
9 - - grammar compiler  
10 -  
11 -SYNOPSIS  
12 - mkdfa.pl [options...] {prefix}  
13 -  
14 -DESCRIPTION  
15 - mkdfa.pl compiles the Julian format grammar (.grammar and .voca) to  
16 - Julian native formats (.dfa and .dict). In addition, ".term" will be  
17 - also generated that stores correspondence of category ID used in the  
18 - output files to the source category name.  
19 -  
20 -  
21 - prefix should be the common file name prefix of ".grammar" and "voca"  
22 - file. From prefix.grammar and prefix.voca file, prefix.dfa, prefix.dict  
23 - and prefix.term will be output.  
24 -  
25 -OPTIONS  
26 - -n  
27 - Not process dictionary. You can only convert .grammar file to .dfa  
28 - file without .voca file.  
29 -  
30 -ENVIRONMENT VARIABLES  
31 - TMP or TEMP  
32 - Set directory to store temporal file. If not specified, one of them  
33 - on the following list will be used: /tmp, /var/tmp, /WINDOWS/Temp,  
34 - /WINNT/Temp.  
35 -  
36 -EXAMPLES  
37 - Convert a grammar foo.grammar and foo.voca to foo.dfa, foo.voca and  
38 - foo.term.  
39 -  
40 -SEE ALSO  
41 - julius ( 1 ) ,  
42 - generate ( 1 ) ,  
43 - nextword ( 1 ) ,  
44 - accept_check ( 1 ) ,  
45 - dfa_minimize ( 1 )  
46 -  
47 -DIAGNOSTICS  
48 - mkdfa.pl invokes mkfa and dfa_minimize internally. They should be  
49 - placed at the same directory as mkdfa.pl.  
50 -  
51 -COPYRIGHT  
52 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
53 -  
54 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
55 -  
56 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
57 - Technology  
58 -  
59 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
60 - Technology  
61 -  
62 -LICENSE  
63 - The same as Julius.  
64 -  
65 -  
66 -  
67 - 10/02/2008 MKDFA.PL(1)  
recognize/src/julius/doc/manuals/mkgshmm.txt
View file @e1cf2b2
@@ -1,50 +0,0 @@ @@ -1,50 +0,0 @@
1 - mkgshmm  
2 -  
3 -MKGSHMM(1) MKGSHMM(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - mkgshmm  
9 - - convert monophone HMM to GS HMM for Julius  
10 -  
11 -SYNOPSIS  
12 - mkgshmm {monophone_hmmdefs}  
13 - >  
14 - {outputfile}  
15 -  
16 -DESCRIPTION  
17 - mkgshmm converts monophone HMM definition file in HTK format into a  
18 - special format for Gaussian Mixture Selection (GMS) in Julius.  
19 -  
20 - GMS is an algorithm to reduce the amount of acoustic computation with  
21 - triphone HMM, by pre-selection of promising gaussian mixtures using  
22 - likelihoods of corresponding monophone mixtures.  
23 -  
24 -EXAMPLES  
25 - (1) Prepare a monophone model which was trained by the same corpus as  
26 - target triphone model.  
27 -  
28 - (2) Convert the monophone model using mkgshmm.  
29 - (3) Specify the output file in Julius with option "-gshmm"  
30 -  
31 -SEE ALSO  
32 - julius ( 1 )  
33 -  
34 -COPYRIGHT  
35 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
36 -  
37 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
38 -  
39 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
40 - Technology  
41 -  
42 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
43 - Technology  
44 -  
45 -LICENSE  
46 - The same as Julius.  
47 -  
48 -  
49 -  
50 - 10/02/2008 MKGSHMM(1)  
recognize/src/julius/doc/manuals/mkss.txt
View file @e1cf2b2
@@ -1,55 +0,0 @@ @@ -1,55 +0,0 @@
1 - mkss  
2 -  
3 -MKSS(1) MKSS(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - mkss  
9 - - calculate average spectrum for spectral subtraction  
10 -  
11 -SYNOPSIS  
12 - mkss [options...] {filename}  
13 -  
14 -DESCRIPTION  
15 - mkss is a tool to estimate noise spectrum for spectral subtraction on  
16 - Julius. It reads a few seconds of sound data from microphone input,  
17 - calculate the average spectrum and save it to a file. The output file  
18 - can be used as a noise spectrum data in Julius (option "-ssload").  
19 -  
20 - The recording will start immediately after startup. Sampling format is  
21 - 16bit, monoral. If outpue file already exist, it will be overridden.  
22 -  
23 -OPTIONS  
24 - -freq Hz  
25 - Sampling frequency in Hz (default: 16,000)  
26 -  
27 - -len msec  
28 - capture length in milliseconds (default: 3000)  
29 -  
30 - -fsize sample_num  
31 - frame size in number of samples (default: 400)  
32 -  
33 - -fshift sample_num  
34 - frame shift in number of samples (default: 160)  
35 -  
36 -SEE ALSO  
37 - julius ( 1 )  
38 -  
39 -COPYRIGHT  
40 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
41 -  
42 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
43 -  
44 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
45 - Technology  
46 -  
47 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
48 - Technology  
49 -  
50 -LICENSE  
51 - The same as Julius.  
52 -  
53 -  
54 -  
55 - 10/02/2008 MKSS(1)  
recognize/src/julius/doc/manuals/nextword.txt
View file @e1cf2b2
@@ -1,82 +0,0 @@ @@ -1,82 +0,0 @@
1 - nextword  
2 -  
3 -NEXTWORD(1) NEXTWORD(1)  
4 -  
5 -  
6 -  
7 -NAME  
8 - nextword  
9 - - display next predicted words (in reverse order)  
10 -  
11 -SYNOPSIS  
12 - nextword [-t] [-r] [-s spname] [-v] {prefix}  
13 -  
14 -DESCRIPTION  
15 - Given a partial (part of) sentence from the end, it outputs the next  
16 - words allowed in the specified grammar.  
17 -  
18 -  
19 - .dfa, .dict and .term files are needed to execute. They can be  
20 - generated from .grammar and .voca file by mkdfa.pl.  
21 -  
22 - Please note that the latter part of sentence should be given, since the  
23 - main 2nd pass does a right-to-left parsing.  
24 -  
25 -OPTIONS  
26 - -t  
27 - Input / Output in category name. (default: word)  
28 -  
29 - -r  
30 - Enter in reverse order  
31 -  
32 - -s spname  
33 - the name string of short-pause word to be supressed (default: "sp")  
34 -  
35 - -v  
36 - Debug output.  
37 -  
38 -EXAMPLES  
39 - Exmple output of a sample grammar "fruit":  
40 -  
41 - % nextword fruit  
42 - Stat: init_voca: read 36 words  
43 - Reading in term file (optional)...done  
44 - 15 categories, 36 words  
45 - DFA has 26 nodes and 42 arcs  
46 - -----  
47 - command completion is disabled  
48 - -----  
49 - wseq > A BANANA </s>  
50 - [wseq: A BANANA </s>]  
51 - [cate: (NUM_1|NUM_1|A|A) FRUIT_SINGULAR NS_E]  
52 - PREDICTED CATEGORIES/WORDS:  
53 - NS_B (<s> )  
54 - HAVE (HAVE )  
55 - WANT (WANT )  
56 - NS_B (<s> )  
57 - HAVE (HAVE )  
58 - WANT (WANT )  
59 -  
60 -  
61 -SEE ALSO  
62 - mkdfa.pl ( 1 ) ,  
63 - generate ( 1 ) ,  
64 - accept_check ( 1 )  
65 -  
66 -COPYRIGHT  
67 - Copyright (c) 1997-2000 Information-technology Promotion Agency, Japan  
68 -  
69 - Copyright (c) 1991-2008 Kawahara Lab., Kyoto University  
70 -  
71 - Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and  
72 - Technology  
73 -  
74 - Copyright (c) 2005-2008 Julius project team, Nagoya Institute of  
75 - Technology  
76 -  
77 -LICENSE  
78 - The same as Julius.  
79 -  
80 -  
81 -  
82 - 10/02/2008 NEXTWORD(1)  
recognize/src/julius/install/julius-4.2.2.tar.gz
View file @e1cf2b2
No preview for this file type