[FFmpeg-devel] [PATCH 2/2] libavformat/dsfdec: Error dsf_read_header fixed.

[Peter Ross]

On Fri, Jul 31, 2015 at 08:56:49PM -0400, Ganesh Ajjanagadde wrote:
> On Fri, Jul 31, 2015 at 8:08 PM, Michael Niedermayer
> <michael at niedermayer.cc> wrote:
> > On Fri, Jul 31, 2015 at 07:33:01PM -0400, Ganesh Ajjanagadde wrote:
> >> On Fri, Jul 31, 2015 at 7:01 PM, Ihar A. Tumashyk <itumashyk at gmail.com> wrote:
> >> > Sample rate is written "as is" in header . Is *should not* be devined by
> >> > 8. Refer spec.:
> >> > http://dsd-guide.com/sites/default/files/white-
> >> > papers/DSFFileFormatSpec_E.pdf
> >> >
> >> > After this patch ffprobe will corretly show sample rate for DSF files.
> >> >
> >> > Signed-off-by: Ihar A. Tumashyk <itumashyk at gmail.com>
> >> > ---
> >> >  libavformat/dsfdec.c | 2 +-
> >> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >> >
> >> > diff --git a/libavformat/dsfdec.c b/libavformat/dsfdec.c
> >> > index ae198b2..3e162ae 100644
> >> > --- a/libavformat/dsfdec.c
> >> > +++ b/libavformat/dsfdec.c
> >> > @@ -105,7 +105,7 @@ static int dsf_read_header(AVFormatContext *s)
> >> >
> >> >      st->codec->codec_type   = AVMEDIA_TYPE_AUDIO;
> >> >      st->codec->channels     = avio_rl32(pb);
> >> > -    st->codec->sample_rate  = avio_rl32(pb) / 8;
> >> > +    st->codec->sample_rate  = avio_rl32(pb);
> >> >
> >> >      switch(avio_rl32(pb)) {
> >> >      case 1: st->codec->codec_id = AV_CODEC_ID_DSD_LSBF_PLANAR; break;
> >>
> >> LGTM, thanks for clarifying intent of previous patch.
> >> I am a little puzzled as to why this was never caught till now,
> >
> > Changing the sample rate like in the patch breaks playback (it plays
> > much too quick)
> > you can find samples to test at:
> > http://www.2l.no/hires/
> 
> I really need to stop trusting these spec files that much,
> they are often incomplete as in this case. Ironically,
> wikipedia here gives a much better description of the format:
> https://en.wikipedia.org/wiki/Direct_Stream_Digital,
> where they mention that essentially this signal is oversampled by a factor of 8.
> As the wiki entry indicates, max frequency of PCM is 352.8 kHz.
> 
> In fact, I think previous patch should be reverted:
> the wikipedia article says that newer format is called "DSD-wide" and
> uses 8-bit word length.
> ffprobe yields the downsampled frequency of 352.8 kHz which is what
> actually gets used when one demuxes/
> operates on the data per wikipedia.
> Thus, I believe previous behavior was correct.

converting between DSD->PCM reduces the sample rate by 8.
at present, the DSD->PCM conversion happens libavcodec/dsddec.c.
however, a 'codec' cannot change sample_rate, so the divide by 8 is performed earlier in libavformat.

yes, this is a hack.

i have a patch set that moves DSD->PCM conversion into swresample, where it belongs, 
along with adding a DST decoder and support for additional file formats.
If anyone is interested, I can repost.

-- Peter
(A907 E02F A6E5 0CD2 34CD 20D2 6760 79C5 AC40 DD6B)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20150801/b0214022/attachment.sig>