[FFmpeg-devel] [PATCH] Bonk decoder and demuxer

Piotr Bandurski ami_stuff at o2.pl
Fri May 6 14:18:41 CEST 2016


> > > On 5/5/16, Piotr Bandurski <ami_stuff at o2.pl> wrote:
> > >> > > > > > > > > Improved version attached.
> > >> > > >
> > >> > > > I found another bug, this time in lossy mode - the file encoded as
> > >> > > > "joint stereo" decodes after a few seconds into noise:
> > >> > > >
> > >> > > > https://www.datafilehost.com/d/c2e8b332
> > >> > >
> > >> > > here are a few more samples:
> > >> > >
> > >> > > https://www.datafilehost.com/d/e4825eb4
> > >> > >
> > >> > > quantization: 1.15
> > >> > > downsampling ratio: 1:1/2:1/3:1/10:1
> > >> > > predictor size: 282
> > >> > > enable join stereo
> > >> > >
> > >> > >
> > >> > > in file "down_2_1.bonk" noise stars around 2:30
> > >> >
> > >> > hmm it has actually not much to do with "join stereo" option, because
> > >> > the same problem happens with mono files:
> > >> >
> > >> > https://www.datafilehost.com/d/09128a25
> > >>
> > >> these mono files hangs with ffmpeg's bonk decoder:
> > >>
> > >> https://www.datafilehost.com/d/b94fd994
> > >
> > > there is actually no hang, I used wrong binary for the test (with applied v1
> > > version of your patch, so it's fixed)
> > 
> > I fixed hangs/crashes but cant locate noise bug anywhere.
> 
> It must be a bug in fre:ac implementation - I located another bug with lossless 6khz file with predictor set to 512
> fre:ac file decodes into noise with bonk and ffmpeg.
> 
> I created similar sample with bonk (-l -s 512) and it decodes correctly with ffmpeg.
> 
> There is also some unrelated thing - bonk encoder doesn't like wav files generated by ffmpeg, but I don't know if this
> is a bug in ffmpeg or bonk.
> 
> so summarizing - LGTM from me for your decoder.

I was able to isolate this and it looks like it's really a bug in encoder (see "bonkdec_bug_qualization_1_15_down_10_1_pred_282.wav")
just original decoder is able to better handle this

https://www.datafilehost.com/d/8ca9e3c1

qualization 1.15
downsampling 10:1
predictor ...



More information about the ffmpeg-devel mailing list