HNM-speech anlysis/synthsis model

a routine for HNM model

Description

The following zip file contains two routines for analysis/synthesis of HNM.
HNM is a analysis/synthesis model of speech,like classical LPC model.
Due to the limitation of my ability,i only implement a simply version of HNM,so welcome for somebody to improve on it.

Usage:
analysis: [p,c,o]=AnHNM('3.wav');
synthesis: s=SyHNM(p,c);

then 'soundsc(s)',enjoy it:-)

Note:special for 8KHz speech in .wav format

• MATLAB release: MATLAB 7.4 (R2007a)

[firas saleh]

 

[q tahmina]

 

[Talal]

Hi!

I would like to ask two questions

On page 32 (on adobe reader 49/225) of HNM thesis (Stylianou) it states that after estimating the pitch by minimizing E the gross
pitch errors are removed (reference to Multiband Excited Vocoder).

I have read that paper and cannot understand where we get tracks from> Before this step I have got the pitch for every frame of the speech signal
So as such there is only one track.

Can you please clarify?

On page 33 (on adobe reader 50/225) for voiced/unvoiced decision a synthetic signal s ̃(t) is generated
using the amplitudes and phases determined by the DFT algorithm.

Can you please specify which algorithm is this? Any papers on it?

I would really appreciate if you can help me!