Question

I trained my Acoustic model and received acceptable accurate rate (85%) over a small data (10 Vietnamese words). but when I integrate this model in to Transcriber sample (packaged with Sphinx 4) and try to transcribe a word, which is in the 10 above words, I received a list of "Falling back to non-recursive partition". I need it to output a word instead.

Here is my project.

Any idea? Thanks

Was it helpful?

Solution

acceptable accurate rate (85%) over a small data (10 Vietnamese words).

This error rate is too high, you need to work on improvement of this result. Accuracy on 10 words must be over 95%. See CMUSphinx troubleshooting section in acoustic model training tutorial for more information.

received a list of "Falling back to non-recursive partition".

This error is a sign of a problem in your input data.

In latest code this error should not be displayed, please also update to a latest version.

Licensed under: CC-BY-SA with attribution
Not affiliated with StackOverflow
scroll top