[hts-users:02899] Re: f0 extraction problem

[Matt Shannon <sms46@xxxxxxxxx>]

On 19/06/11 16:36, chimi chimi2011 wrote:
> 1.the mgc extraction is finished by SPTK tools.However, the total frame 
> number of mgc is different from the f0 . why? I think they should keep 
> exactly the same.

I've been slightly bothered by that too, but I don't think I ever
tracked down the cause. For what it's worth, for the non-STRAIGHT demo
lf0 is correct, and mgc always has 2 too few frames. For the STRAIGHT
demo, mgc always has the correct number of frames, and lf0 and bap
usually have the correct number of frames but sometimes (48 of 1132
utts) have 1 too many frames.

During the initial training phase these small inconsistencies are
swamped by a huge inconsistency with the provided label files, which
quite often (169 of 1132 utts) are short by 9 frames or more, and are
sometimes short by as much as 57 frames (cmu_us_arctic_slt_b0340).

Matt