Formant modeling and smoothness estimation

Modeling is relatively basic. More complicated methods may be added later.

The following process is carried out for each of J formants:

1. Time is normalized so that it goes from 0 to 1.

2. Cosine predictors are generated using normalized time. By default up to 5th order predictors are used. This means we can model formant contours faithfully even if they are somewhat wiggly.

3. The formant track is predicted using an OLS regression analysis.

 

The output of this process is:

• A set of regression coefficients for each formant, for each vowel.

• The above can be used to generate predicted formant values for each track.