Sphinx2

Notes on Sphinx2

Creating Language Models

 How to build the language model/dictionary for sphinx2:
    http://www.speech.cs.cmu.edu/sphinx/doc/sphinx-FAQ.html

    Sphinx2 only needs a pronounciation dictionary (.dict), and a language model (.lm)

 Building a dictionary:
 http://www.speech.cs.cmu.edu/cgi-bin/cmudict
    ftp://ftp.cs.cmu.edu/afs/cs.cmu.edu/data/anonftp/project/fgdata/dict/
       ftp://ftp.cs.cmu.edu/afs/cs.cmu.edu/data/anonftp/project/fgdata/dict/cmudict.0.6.gz

       You would grep the dictionary for your words
       Strip out the stress numbers

 Building language models
 http://www.speech.cs.cmu.edu/SLM/toolkit.html
    http://www.speech.cs.cmu.edu/SLM/CMU-Cam_Toolkit_v2.tar.gz

 cat output/zork.corpus | ./text2wfreq > output/zork.wfreq
 cat output/zork.wfreq | ./wfreq2vocab -top 20000 > output/zork.vocab
 cat output/zork.corpus | ./text2idngram -vocab output/zork.vocab > output/zork.idngram
 ./idngram2lm -idngram output/zork.idngram -vocab output/zork.vocab -binary output/zork.binlm

 ./idngram2lm -idngram output/zork.idngram -vocab output/zork.vocab -arpa output/zork.lm

last edited September 23, 2006 — ( info ) ( diff )