silan < 0.3.3 cueout wildly off in some cases after uploading files

[leonardpg]

Just started with LibreTime. I uploaded maybe 50 files and I've discovered at least 5 that have incorrect cueout times after upload. Sometimes up to a minute or more. I discovered it when playing back a test of a smartblock.

Is there someway to tweak the analyzer so that this does not happen? Otherwise, I was thinking of erring on the side of dead air and running a query that sets cueout = length.

Thanks!

[Robbt]

I'm figuring this must be an issue with the Silan silence detection. Are you running on Ubuntu 14.04 ? I haven't looked into whether airtime analyzer sets cueouts based upon Silan. Is there any silence in these tracks that might be causing the issue or is it seemingly random ?

I used to see this in the good-old days when I had incoming variable bitrate VBR files. I now preprocess those with SOX to CBR

[hairmare]

Analyzer uses silan to detect cue points. I think this is fixed with silan >=0.3.3. Analyzer already has some code to ignore erroneous values from silan but it's impossible for it to be perfect.

I already opened an NMU for debian https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=855319 to get it updated but it needs a new maintainer. I think you can remove silan completely to get it to not do anything but haven't tested that.

[Robbt]

If and when I have time I'm interested in helping with the maintenance of the Debian packages that are currently unmaintained, but I need to spend the time figuring out the process for doing so again.

[leonardpg]

Uploaded a bunch more last night...What seems to be happening in some cases is the cueout and length are BOTH being set to the same value. It doesn't seem to be specific to VBR, and when I listen to the point in the song where the cueout and length are set, it's not silent. So...any ideas?

Edit: I looked at the database at one of the effected files and I see this:
airtime=# select cuein,cueout,length from cc_files where artist_name = 'Hurray for the Riff Raff';
cuein | cueout | length
-----------------+-----------------+-----------------
00:00:00.256372 | 00:01:34.982698 | 00:03:18.466667

Correct info for length, wrong cueout. However, when the file comes up in the UI, I see this:

cue out: 01:35 length: 01:34.7

Are they stereo?

[leonardpg]

Yes, stereo.
I'm getting an error when I try to build the latest version of silan from source.

What's the way, as suggested above, to just remove the cuepoint analyzer from the analyzer process?

[hairmare]

I think you can apt-get remove silan. I might just fail even worse though.

If building silan locally fails you might need to install some build deps. I think libavformat-dev, libavcodec-dev, libavutil-dev, libsndfile1-dev, and pkg-config should cover what's needed.

I fixed this in Ubuntu Trusty here: #197

[hairmare]

If you are on Debian or Ubuntu you can check if the NMU packages from the debian bugreport work. I built them on OBS so we can test them. Please keep in mind that these are experimental, I did not check to see if they will be upgradeable once the debian patch is merged

Refer to https://github.com/LibreTime/libretime/releases for current instructions.

You can check that you got the nmu build with apt-cache show silan where the second line should display 0.3.3~nmu1 as version or by running silan --version.

I couldn't get the packages built for Ubuntu Trusty. I was able to satisfy all distros build requirements except Ubuntu Trusty, the libav packages silan needs are missing from OBS. This seems to be a result of the libav/ffmpeg struggles. I found a universe libav package on OBS to build against that should work and updated the example above with an Ubuntu Trusty example.

The OBS packages are currently on my home repo in OBS, I'll get them moved to wherever we focus on with the LibreTime packages. Feel free to fork em on OBS if you would like to give building them for Trusty a crack.

Update I updated the package name to use ~nmu1 instead of +nmu1 to allow for an easy update once the Debian version is bumped. Users that already installed the +nmu1 might need to look out for this if silan gets bumped upstream.