Software

From SpeechWiki

(Difference between revisions)
Jump to: navigation, search
m
m (style)
Line 1: Line 1:
=Statistical Speech Technology Group Software=
=Statistical Speech Technology Group Software=
-
Our policy: everything we write is free on the web. This wiki page is intended to be definitive, because anybody in the group can edit it to add their own software. http://www.isle.uiuc.edu/software/index.html provides a spider-indexable backup copy.
+
Our policy: everything we write is free on the web. This wiki is intended to be definitive, because anybody in the group can edit it to add their own software. A spider-indexable backup is at http://www.isle.uiuc.edu/software .
-
All of our software is available via [http://subversion.tigris.org subversion], using login name "anon" with no password (hit "enter" when a password is requested).
+
Our software is available via [http://subversion.tigris.org subversion], using login name "anon" with no password (hit "enter" when a password is requested).
-
* If you are using Windows, download [http://tortoisesvn.tigris.org/ TortoiseSVN]
+
* On Windows, download [http://tortoisesvn.tigris.org/ TortoiseSVN]
-
* If you are using linux, use the svn command interface, e.g., svn co svn://mickey.ifp.uiuc.edu/speechfileformats
+
* On Linux, use the svn command, e.g., svn co svn://mickey.ifp.uiuc.edu/speechfileformats
<table border=2><tr>
<table border=2><tr>
Line 24: Line 24:
<tr><td>Signal Processing</td></tr>
<tr><td>Signal Processing</td></tr>
-
<tr><td>PVTK</td><td>Extract HTK features as training vecs for libSVM, apply trained SVMs directly to feature files<br>
+
<tr><td>PVTK</td><td>Extract HTK features as training vectors for libSVM, apply trained SVMs directly to feature files<br>
[http://www.isle.uiuc.edu/software/PVTK2005May23.tgz TGZ archive], [svn://mickey.ifp.uiuc.edu/PVTK SVN repository] (Sarah Borys and MH 2005-8)
[http://www.isle.uiuc.edu/software/PVTK2005May23.tgz TGZ archive], [svn://mickey.ifp.uiuc.edu/PVTK SVN repository] (Sarah Borys and MH 2005-8)
</td></tr>
</td></tr>
-
<tr><td>VAD</td><td>Voice activity detector w/improved noise model<br>
+
<tr><td>VAD</td><td>Voice activity detector with improved noise model<br>
[http://www.isle.uiuc.edu/pubs/2007/lee07dspincars.pdf Description],  
[http://www.isle.uiuc.edu/pubs/2007/lee07dspincars.pdf Description],  
[http://www.isle.uiuc.edu/software/lee_vad.m lee_vad.m], [svn://mickey.ifp.uiuc.edu/lee_vad SVN repository] (Bowon Lee, 2007)</td></tr>
[http://www.isle.uiuc.edu/software/lee_vad.m lee_vad.m], [svn://mickey.ifp.uiuc.edu/lee_vad SVN repository] (Bowon Lee, 2007)</td></tr>
Line 62: Line 62:
<tr><td>improved MVA</td><td>Perform mean and variance normalization and ARMA filtering<br>
<tr><td>improved MVA</td><td>Perform mean and variance normalization and ARMA filtering<br>
-
It's essentially the version found [http://ssli.ee.washington.edu/people/chiaping/mva.html here],  
+
It's essentially [http://ssli.ee.washington.edu/people/chiaping/mva.html this] version,  
-
with these improvements:
+
improved:
* better error reporting (e.g. failing to open file tells you so instead of core dumping)
* better error reporting (e.g. failing to open file tells you so instead of core dumping)
* more accurate mean and variance estimation (doubles instead of floats in strategic places)
* more accurate mean and variance estimation (doubles instead of floats in strategic places)

Revision as of 16:29, 4 February 2009

Statistical Speech Technology Group Software

Our policy: everything we write is free on the web. This wiki is intended to be definitive, because anybody in the group can edit it to add their own software. A spider-indexable backup is at http://www.isle.uiuc.edu/software .

Our software is available via subversion, using login name "anon" with no password (hit "enter" when a password is requested).

Learning
PronounceLetters to phones using an HMM

Description,Demo,

SVN archive (Arthur Kantor, 2007)
HDKHTK-based Explicit-duration HMM

Description, TGZ archive, SVN repository (Ken Chen, 2003)

HTKtrainScripts for training HMMs using HTK

SVN repository (Sarah Borys and Mark Hasegawa-Johnson, 2008)

Signal Processing
PVTKExtract HTK features as training vectors for libSVM, apply trained SVMs directly to feature files

TGZ archive, SVN repository (Sarah Borys and MH 2005-8)

VADVoice activity detector with improved noise model

Description,

lee_vad.m, SVN repository (Bowon Lee, 2007)
Computation
GMTK Parallel Split GMTK commands into batch jobs for a cluster

Description,

SVN repository (Arthur Kantor, 2008)
HTK Parallel

Split an HTK command into batch jobs for a cluster (Bowon Lee, 2006)
Description, HCopy.pl, HVite.pl, HERest.pl,

HResults.pl, SVN repository
Data
dtmfsegSegment audio files at DTMF tones
SVN repository (Bowon Lee, 2006)
transcription toolsConvert transcription formats
TGZ archive, SVN repository (Mark Hasegawa-Johnson, 2005)
speechfileformatsRead and write HTK files in matlab
TGZ archive, SVN repository (Mark Hasegawa-Johnson, 2004)
CTMReditManually and automatically segment CT and MR image stacks

Description, SVN repository (Jul Cha and MH 1999)

improved MVAPerform mean and variance normalization and ARMA filtering

It's essentially this version, improved:

  • better error reporting (e.g. failing to open file tells you so instead of core dumping)
  • more accurate mean and variance estimation (doubles instead of floats in strategic places)
  • faster computation in the case of MV (ARMA order 0)

source: svn://mickey.ifp.uiuc.edu/corporaNormalizationScripts/fisher/MVA.cc

binary: http://mickey.ifp.uiuc.edu/speech/akantor/fisher/programs/bin.Linux/MVA

Scriptsmiscellaneous perl, python, bash, and ruby SVN archive, Documentation

Phonetic Transcription Tool

This tool gives the American English phonetic transcription of any string. It uses an HMM model to either generate a most likely phonetic transcription, or if a phonetic transcription is provided, it can perform forced alignment. So, it gives a reasonable pronounciation for out-of-dictionary words, or partially pronounced words.

Phonetic Transcription Tool. You can try out the demo here.

Perl scripts for parallel processing of HTK commands

Perl scripts for parallel processing of four HTK commands (HCopy, HERest, HVite, and HResults) are available.

1. HCopy.pl

2. HERest.pl

3. HVite.pl

4. HResults.pl

Those scripts use the SGE (Sun Grid Engine) for job queuing.

Detailed information about the above Perl scripts can be found here.

Brief introduction about the SGE can be found here


Written by Bowon Lee, 02/24/2006

Personal tools