[Lame-dev] Re: calc_noise bug

[Mark T. <mt...@mp...>]

I had thought this was fixed, but in today's code this
problem is back:


 > I haven't looked into this, but here's an assert failure in
 > the new calc_noise:
 >
 > You can get this 50K .wav file that triggers this
 > at http://www.mp3dev.org/tmp
 >
 >
 > ~/mp3/lame >frontend/lame -h -b 96 calc_noise_bug.wav
 > LAME version 3.93 (alpha 2, May 11 2002 12:09:14) (http://www.mp3dev.org/)
 > Resampling:  input 44.1 kHz  output 32 kHz
 > Using polyphase lowpass  filter, transition band: 12000 Hz - 12387 Hz
 > Encoding calc_noise_bug.wav to calc_noise_bug.wav.mp3
 > Encoding as 32 kHz  96 kbps j-stereo MPEG-1 Layer III (10.7x) qval=2
 >     Frame          |  CPU time/estim | REAL time/estim | play/CPU |    ETA
 >      0/7692   ( 0%)|    0:00/    0:00|    0:00/    0:00|   0.0000x|    0:00
 >
 > lame: quantize_pvt.c:615: calc_noise: Assertion `0 <= s && s < 330' failed.
 > Abort
 
 the global gain is zero in this case. Because the pretab flag is set the 
 resulting stepsize is -2, which is negative and not allowed. (it's happening
 for scalefactor 13)
 
 Seems we have to make sure that the global gain does not fall below
 
 ((scalefac->l[sfb] + (cod_info->preflag ? pretab[sfb] : 0))
 	       << (cod_info->scalefac_scale + 1))
 
 Or scalefactors have to be limited by the global gain.
 
 > Mark
 
 
 Ciao Robert