Michael Knight - 2004-01-29

I am working with the released code (sphinx3-0.1.tar.gz). I am getting very inaccurate results with livepretend.exe.  I was able to get accurate results with the same language and acoustic models from decoder.exe.

To get good results with decoder.exe (compiled under cygwin) I changed the hardwired feature file format from 1s_c_d_dd to s3_1x39 and used a feature file created with wave2feat from the sphinx train release.

I was not able to get accurate results from livepretend.exe(compiled under vs.net) with the feature file format of 1s_c_d_dd  or s3_1x39. I have tried several audio files in several formats. I believe that livepretend wants PCM 16bit 16htz data with no header. I have attempted the process with both header info and with out with no valid results.

Any idea what could be going wrong? Is there another format of input data that I should try?

Thanks in advance,
Michael