Starting Pocket sphinx with small audio of numbers

[abdul rashid]

I am new to pocketsphinx and starting with the latest version on windows. I have audio of numbers which I am trying with pocketsphinx. The audio contains the words (eight eight four one six seven three six nine). but its shows as eighty four one six seven three six nine.
Any help/guidance is appreciated.

[Nickolay V. Shmyrev]

Speech recognition is not always accurate. For more details on how to make it accurate read http://cmusphinx.github.io/wiki/tutorialtuning

[abdul rashid]

I am trying to train model . Its a small audio and I am facing the Error "Failed to Align audio to transcript". Can you help me out to understand it. I am attaching the Audio file , the transcript and error deatails in text file.

[Nickolay V. Shmyrev]

You miss </s> in the x.transcription file before (x)

[abdul rashid]

Thanks . I added at the end but still it say "Failed to Align audio to transcript". I have attached the log.

[Nickolay V. Shmyrev]

Your audio file is 8khz, you are trying to extract features with samprate 16000 and probably using 16-khz model, this is not a good idea. The command line is also different from tutorial, you miss argfile argument in sphinx_fe.

[abdul rashid]

I changed audio to 16khz. Also i am using -argfile en-us/feat.params as argument in sphinx_fe. Still having same error.

[Nickolay V. Shmyrev]

You can not do that, you need to use 8khz model and follow the tutorial properly.