Regression: Encoding Vorbis from FLAC source causes audible artefacts at regular intervals

[Anton Eliasson]

Binary package hint: soundconverter

When I convert FLAC files to Ogg Vorbis using Soundconverter, audible artefacts are introduced at every ~40,5 seconds in the form of short breaks in the stream. These are slightly noticeable in Mplayer (1.0rc4-4.5.2) and very noticeable in Totem (2.32.0). They also cause my Sandisk Sansa Clip+ portable music player to pause and skip to the next track.
So far I've only found this problem to occur when encoding FLACs directly to Vorbis. I've been able to reproduce it with several different source files from different artists and albums, originally encoded by both me and others. It does not occur if I first decode the FLAC to WAVE using 'flac -d <file.flac>' and then compress the WAVE file to Vorbis. Nor does it occur if I convert the same FLAC source to MP3.

Here are a few files encoded with different bitrates:
http://www.nastalektion.se/dump/01-Hunab_Ku-64k.ogg
http://www.nastalektion.se/dump/01-Hunab_Ku-128k.ogg
http://www.nastalektion.se/dump/01-Hunab_Ku-256k.ogg
and the source:
http://www.nastalektion.se/dump/01-Hunab_Ku.flac

I'm fairly sure that this bug is a regression. Here's a file I encoded with Soundconverter a year ago, without audible artefacts:
http://www.nastalektion.se/dump/01_-_Don_t_Tell_Me_That_It_s_Over-20100614.ogg
Here's one I encoded today with more or less the same settings, from the same source. This has the choppiness every 40,5 seconds I mentioned earlier:
http://www.nastalektion.se/dump/01_-_Don_t_Tell_Me_That_It_s_Over-20110613.ogg
And the source:
http://www.nastalektion.se/dump/01%20-%20Don't%20Tell%20Me%20That%20It's%20Over.flac

Please note that all the files I've linked to are copyrighted to their respective owners. Only use them for educational purposes, such as for solving this bug. Please don't spread them on.

ProblemType: Bug
DistroRelease: Ubuntu 11.04
Package: soundconverter 1.4.4-2
ProcVersionSignature: Ubuntu 2.6.38-8.42-generic 2.6.38.2
Uname: Linux 2.6.38-8-generic x86_64
NonfreeKernelModules: nvidia
Architecture: amd64
CheckboxSubmission: c43a19076448baad41c3944df9d92f15
CheckboxSystem: 533a8038ef5a55601694b49142442fa7
Date: Mon Jun 13 19:16:41 2011
InstallationMedia: Ubuntu 10.10 "Maverick Meerkat" - Release amd64 (20101007)
PackageArchitecture: all
ProcEnviron:
 LANGUAGE=sv_SE:en
 PATH=(custom, user)
 LANG=sv_SE.utf8
 SHELL=/bin/bash
SourcePackage: soundconverter
UpgradeStatus: Upgraded to natty on 2011-04-28 (46 days ago)

[Anton Eliasson (eliasson)]

I did some troubleshooting today:
This bug is not present in Maverick running as a virtual machine in Virtualbox.
Back to Natty: Converting a flac file as described above as a different user yielded the same result, which should rule out errors caused by defect config files in my home dir. I tried building soundconverter from source code in different variations. First the one distributed with Ubuntu (apt-get source soundconverter; fakeroot debian/rules binary; dpkg -i soundconverter_1.4.4-2_all.deb), the same version from upstream (tar xf soundconverter-1.4.4.tar.bz2; ./configure; sudo checkinstall) and the latest upstream version (tar xf soundconverter-1.5.4.tar.gz; ./configure; sudo checkinstall). All of these were affected by this bug.

This might be a backend issue. The version of Soundconverter shipped with Natty gives this startup message:
SoundConverter 1.4.4
  using Gstreamer version: 0.10.32, Python binding version: 0.10.21
  using gio
  using 4 thread(s)
While in Maverick the startup message is:
SoundConverter 1.4.4
** Message: pygobject_register_sinkfunc is deprecated (GstObject)
  using Gstreamer version: 0.10.30, Python binding version: 0.10.19
  using gio
  'lame' element not found, disabling MP3.
  'faac' element not found, disabling AAC.
  using 2 thread(s)

I'll look around and try to find out how much pain it is to update Gstreamer to 0.10.32 in Maverick.

[Anton Eliasson (eliasson)]

Well, here's what I did: I opened my virtual Maverick's /etc/apt/sources.list up in gedit and replaced all instances of maverick with natty. OH NO YOU DIDN'T, I can hear you say. But hey, I'll just roll back to a previous snapshot when I'm done, no problem. So, after that I ran an apt-get update and an apt-get upgrade but I didn't actually upgrade anything. I just wanted a list of gstreamer packages to upgrade, specifically:
gstreamer0.10-alsa gstreamer0.10-nice gstreamer0.10-gnonlin gstreamer0.10-plugins-base gstreamer0.10-plugins-base-apps gstreamer0.10-plugins-good gstreamer0.10-pulseaudio gstreamer0.10-tools gstreamer0.10-x
apt-get install <the above list> which upgraded what I wanted as well as these packages:
libasound2 libc-bin libc-dev-bin libc6 libc6-dev libgssdp-1.0-2
  libgstfarsight0.10-0 libgstreamer-plugins-base0.10-0 libgstreamer0.10-0
  libgupnp-1.0-3 libjpeg62 libnice10 libnih-dbus1 libnih1 liborc-0.4-0
  libpython2.7 libsqlite3-0 multiarch-support python2.7 python2.7-minimal
I realised that soundconverter wasn't installed at that point so I just pulled it down with apt-get install soundconverter. And look at that:
anton@spire:~$ soundconverter
SoundConverter 1.4.4
** Message: pygobject_register_sinkfunc is deprecated (GstObject)
  using Gstreamer version: 0.10.32, Python binding version: 0.10.19
  using gio
  'lame' element not found, disabling MP3.
  'faac' element not found, disabling AAC.
  using 2 thread(s)
The encode went fine and the resulting file had no stutter every 40,5 s. The next step is to just bite the bullet and do a full apt-get upgrade. Will get back to this bug when that's finished.

[Anton Eliasson (eliasson)]

Updating the system didn't seem to cause any major problems, it survived a reboot just fine. However, this bug never occurred even though both Gstreamer and Soundconverter are of the same version as they are in my native 11.04 installation. So I guess this is just another weird bug caused by updating to Ubuntu Vista 11.04. Since I have absolutely no idea how to troubleshoot this one any further, I'll probably just let it expire.

[Anton Eliasson (eliasson)]

This is an upstream problem. I am now using Debian Wheezy on a different computer and I can still reproduce this. Here's a sample I just made: http://antoneliasson.se/publicdump/01_-_Prelude.ogg, with source: http://antoneliasson.se/publicdump/01_-_Prelude.flac

[Anton Eliasson (eliasson)]

This has sort of already been reported upstream: https://bugs.launchpad.net/ubuntu/+source/soundconverter/+bug/792742, however the conclusion was made that this is instead a problem in GStreamer and not in Soundconverter. I also tested one of my converted files with ogginfo and it did indeed produce a similar output.

See also https://trac.xiph.org/ticket/1807 and https://bugzilla.redhat.com/show_bug.cgi?id=722667. The latter contains a patch (https://bugzilla.redhat.com/attachment.cgi?id=514575) for GStreamer which fixes the bug. By compiling gstreamer0.10-plugins-good from source with this small patch, I can confirm this. Therefore, I am now marking this bug as invalid, since it wasn't in neither Soundconverter nor Ubuntu after all.