I am try to build ASR for Tamil language, i create a small database for Isolated word and centences. i decoded with Sphinx3 decoder, and it was success in isolated word ( 8 - 120% WER). but it seems to
recognize in sentence level it shows 100% WER. i attached result file. please can anyone help me to reduce WER ...
I am try to build ASR for Tamil language, i create a small database for Isolated word and centences. i decoded with Sphinx3 decoder, and it was success in isolated word ( 8 - 120% WER). but it seems to
recognize in sentence level it shows 100% WER. i attached result file. please can anyone help me to reduce WER ...
Also i uploaded my .cfg and .lm file for yor reference follow link: http://www.files.com/set/57e0e246d3963
Last edit: Nickolay V. Shmyrev 2016-09-20
You need to share full model training folder. For sharing it's better to use dropbox or google drive.