SpeakGoodChinese Pronunciation recognition

The pronunciation recognizer works like the SGC tone recognizer, but instead of
tone models, it uses real speech examples of correct and incorrect
pronunciations. The recognizer is build completely in Praat scripts and runs on
praat and sendpraat executables. There is a basic, Tcl/Tk interface for testing
available.

The sound models are generated beforehand, stored in the wav directory and
converted to MFCC analysis files in the mfcc directory. The recognizer uses DTW
to calculate the distance between the speech sample and all the models for that
pinyin word. The model with the shortest distance is considered to be the
recognized word.

There are three possible sources of sound models for recognition. All three are
represented:
1 Record samples of correct and incorrect words from speakers, using shadowing

2 Select native speech samples to represent correct and incorrect pronunciation

3 Generate correct and incorrect samples by TTS

All sound and MFCC directories are two levels deep. The subdirectories are named
after the target word and each contain all sound models, correct and incorrect,
for that target word.

COMPONENTS

Praat executables:
praat4512_win.exe
praatLinux
praatWin.exe
sendpraat
sendpraat.exe
sendpraatLinux
sendpraatWin.exe

Recognizer scripts:
SGC_PronProt.praat           : Main script
SGC_PronProt.tcl             : Tcl/Tk test interface
SoundRecognition.praat       : DTW recognizer

Auxiliary and test scripts:
PlayPronExample.praat        : Play an example recording
DrawToneContour.praat        : Part of the Tone recognizer (delete)
ReplayPronLogFile.praat      : Replay the complete log file to the recognizer
TestPronProt.praat           : Apply the recognizer to a directory of speech
files
Wav2MFCC.praat               : Convert a directory of wav files to MFCC
wav                          : Directory to store wav files, two levels
examples                     : Directory to store examples for the student

Sound models:
mfcc                         : Main directory for shadowed models
native_mfcc                  : Some examples of using native speech models
tts_mfcc                     : Using TTS generated sound models.

Text files:
README.txt                   : This text
ExampleRecordings.txt        : List of examples available
SGC_protWordList.txt         : A wordlist for the SGC_PronProt.tcl interface


Temporary files:
currentSound.wav             : Used by SGC_PronProt.tcl and SGC_PronProt.praat
lastResult.txt               : Recognizer result
_tmprec.mfcc                 : Used by SGC_PronProt.tcl
_tmprec.wav                  : Used by SGC_PronProt.tcl

Log files and directories:
log                          : Store all recorded sounds, if required
logFile.txt                  : Log file of all activity and results