[hts-users:00739] Re: Is it possible to fix corrupted .cmp files

[Keiichi Tokuda <tokuda@xxxxxxxxxxxx>]

From: "Heiga ZEN (Byung Ha CHUN)" <zen@xxxxxxxxxxxxxxx>
Subject: [hts-users:00734] Re: Is it possible to fix corrupted .cmp files
Date: Fri, 29 Jun 2007 20:38:31 +0900
Message-ID: <4684EF37.4000501@xxxxxxxxxxxxxxx>

> If all values in a frame are 0.0, extracted mel-cepstral coefficients become NaN.

Let me give some additional explanation

By setting a very small value (floor for the FFT spectrum) to -e
option of mcep, the problem can be avoided.  Unfortunately, -e
option has a bug (in _mcep.c):

<         x[i] = x[i]*x[i] + y[i]*y[i];
<         c[i] = log(x[i]+e);
---
>         x[i] = x[i]*x[i] + y[i]*y[i] + e;
>         c[i] = log(x[i]);

accordingly it does not work properly :-(

This will be fixed in the next release of SPTK.

Regards,
Keiichi Tokuda
tokuda@xxxxxxxxxxxx
http://www.sp.nitech.ac.jp/~tokuda/