Try smart vowel extraction
fulldecent opened this issue · comments
Some original research to do.
Make a model for vowel utterances which expects:
- Three formants in 500 Hz - 3000 Hz range
- Little change in formants for a minimum of (1/10 sec?)
- LPC model size ~10
- Model error (consonants, other stuff) below a certain threshold
Then check which parts of the signal meet this model requirements. That is the vowel.