[hts-users:00737] Re: Is it possible to fix corrupted .cmp files

["Nicholas Volk" <nvolk@xxxxxxxxxx>]

Hi,


> Hi,
>
> Nicholas Volk wrote (2007/07/04 18:30):
>
>> But no matter what I do, I still get the same error +2126 when trying
>> to a 44100Hz voice, which is the original sample rate of my speech
>> database. (Same data fine when downsampled to 8000, 16000 and 22100.)
>> However this time there are now "nans" anywhere.
>
> Did you modify the following variables for 44100Hz sampling?
>
> SAMPFREQ    = 16000
> FRAMELEN    = 400
> FRAMESHIFT  = 80
> FFTLEN      = 512
> FREQWARP    = 0.42
> MCPORDER    = 24
>

I'm not wholly sticking to the original data preparation scripts anymore,
since I do labs and f0 stuff with other tools.
I use mkdata.pl to create mcep and cmp files:

mkdata.pl:
----------

# speech analysis setting ===============
$sampfreq    = 44100;  # 16kHz sampling frequency
$framelength = 0.025;  # 25ms window length
$frameshift  = 0.005;  # 5ms frame shift
$windowtype  = 0;      # window type
$normtype    = 1;      # normalization of window
$FFTLength   = 512;    # FFT length for Mel-cepstral analysis
$freqwarp    = 0.42;   # frequency warping for mel
if ( $sampfreq == 8000 ) { $freqwarp = 0.31; }
if ( $sampfreq == 22050 ) { $freqwarp = 0.45; }
if ( $sampfreq == 44100 ) { $freqwarp = 0.45; }
$mceporder   = 28;     # order of Mel-cepstral analysis


In the actual training I use following:

Config.pm:
----------

%ordr = ('mcp' => 29, # feature order (NV)
         'lf0' => '1');

# Speech Analysis/Synthesis Setting ==============
# analysis
$sr = 44100;   # sampling rate (Hz)
$fs = $sr/200;      # frame period (point) (NV)
$fw = 0.42;    # frequency warping
if ( $sr == 8000 ) { $fw = 0.31; }
if ( $sr == 22050 ) { $fw = 0.45; }
if ( $sr == 44100 ) { $fw = 0.45; }
$fr = $fs/$sr; # frame period (sec)

br,
  Nicholas