[hts-users:01640] Re: HTS_vocoder.c: float to short conversion

[Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>]

Hi, Nicholas

Thank you for your cooperation.
I'll release a next version of hts_engine API with these replacement and 
bug fix.

Regards,
Keiichiro Oura

Nicholas Volk wrote:
> Hi,
>
> Some minor corrections/suggestions...
>
> The conversion of sample from float to short x in HTS_vocoder.c is
> currently done as follows:
>
>       /* output */
>       xs = (short) x;
>
> However, if x exceeds the minimum or maximum value of short, the
> signedness may go wrong (you can see the difference in alice01.wav when
> using LSP (Cygwin/gcc)). A simple fix should do:
>
>       /* output */
>       if ( x > 32767 ) { x = 32767; }
>       else if ( x < -32768 ) { x = -32768; }
>       xs = (short) x;
>
>
> Also in the same file, I think it would be much more simpler to replace
> v->pade with:
>
> static const double pade[21] = {
>   1.00000000000,
>   1.00000000000, 0.00000000000,
>   1.00000000000, 0.00000000000, 0.00000000000,
>   1.00000000000, 0.00000000000, 0.00000000000, 0.00000000000,
>   1.00000000000, 0.49992730000, 0.10670050000, 0.01170221000, 0.00056562790,
>   1.00000000000, 0.49993910000, 0.11070980000, 0.01369984000, 0.00095648530,
>   0.00003041721
> };
>
> (Less code, no memory allocations etc required...)
>
> br,
>   Nicholas

Nicholas Volk wrote:
> HTS_PStreamSet_create() has a minor bug:
>
>                         if (not_bound || k == 0)
>                            pst->sm.ivar[msd_frame][m] =
>                                HTS_finv(HTS_SStreamSet_get_vari
>                                         (sss, i, state, m));
>                         else
>                            pst->sm.ivar[msd_frame][k] = 0.0;
>
> The last line should be
>
>                            pst->sm.ivar[msd_frame][m] = 0.0;
>
>
> In HTS_demo-SLT m == k, so the bug is harmless there, but it might
> theoretically affect other implementations.
>
> br,
>   Nicholas