[hts-users:04177] Re: Aperiodicity bands in HTS demo v2.2 @ 48kHz

[Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>]

Hi,

I don't remember the implementation.
Anyway, could you check following URL?

http://data.cstr.ed.ac.uk/emime/HTS-ParallelProcessing-EMIME-ver0.5.tar.gz
http://www.festvox.org/blizzard/bc2010/CSTR_Blizzard2010.pdf

Regards,
Keiichiro Oura

2014-11-29 21:07 GMT+09:00 Daniel Erro <daniel.erro@xxxxxx>:
> Hi all!
> Can anybody tell me what is the criterion to determine the boundaries of the
> STRAIGHT aperiodicity bands in HTS demo v2.2 for 48kHz sampling freq?
> Apparently, according to some docs I have found, the bands are obtained as
> follows (Matlab code):
>
> Nbins=1024;                                          % number of points
> (bins) of the ap envelope
> fs=48000;                                            % sampling freq
> ff=(fs/2)*(0:Nbins)/Nbins;                           % exact frequencies of
> the STRAIGHT bins
> bb=26.81*ff./(1960+ff)-0.53;                         % Hz to bark
> bb=bb+0.15*(bb<2).*(2-bb)+0.22*(bb>20.1).*(bb-20.1); % corrective terms in
> [Traunmueller, JASA 1990]
> bands=round(bb);                                     % to integer
>
> However, the resulting number of bins inside each band is slightly different
> from what I see in data/Makefile:
>
> Mine: 3 4 4 4 5 5 5 6 7 7 9 10 11 12 15 18 21 26 33 43 52 62 87 129 210 237
> Demo: 2 3 4 4 4 5 5 6 6 7 8  9 10 12 14 16 19 24 29 37 49 68 99 160 300 125
>
> By eliminating the corrective term and the -0.53 term I almost got it!
>
> Mine: 2 3 4 4 4 5 5 6 6 7 8 9 11 11 14 16 19 24 29 37 48 67 98 157 293 138
> Demo: 2 3 4 4 4 5 5 6 6 7 8 9 10 12 14 16 19 24 29 37 49 68 99 160 300 125
>
> I have tried many other combinations (removing terms, using alternative bark
> scale formulations, using floor() or ceil() instead of round(), etc) with no
> success. Which formula should I use to get the exact band distribution?
> Thank you very much!!
>
> Daniel Erro
>
> Ikerbasque Research Fellow @ Aholab, UPV/EHU
>
>
>