SoundConverter 1.5.0

Haaaa. Finally a few days of full time work on SoundConverter, so I can launch the 1.5.0 version.
We now have a fully asynchronous core. The old task manager was polling, a reminisent of good old gstreamer 0.8 era, causing tasks faster than 10ms to wait. And the results are awesome, just take a look at my little benchmark bellow:


And that's only numbers. Try it to see the difference. As a bonus, multithreading is also much faster with the new version. And don't ask for the 1.4.4 results with 16GB of sound files, the test wasn't able to finish without crashing.

While investigating on a crash with a friendly user, I remarked the huge amount of memory used by SoundConverter. And just to load tags! I never used it with lots of files at the same time, but this had to be fixed. I tracked down every byte allocated to remove all the unnecessary ones. To my surprise, I was storing all tags passed by gstreamer (even embedded pictures), when we only use a few ones to build filenames, that alone saved a ton of memory. Anyway there it is. Using 3x less memory and 6x faster to load files. Happy converting!

Comments

Post a Comment

Popular posts from this blog

Choosing a CRT-Shader for Youtube

Image
This is part two of the capture of  Stars for Youtube. Video Generation The process is rather simple: Run demo on Dosbox, capture the video. You get a 70fps 320x200 video using the lossless zmbv codec for images, and lossless PCM audio. As the demo is a plain VGA program, this is a perfect copy of what is displayed/played by the demo. Play the 320x200 video with Retroarch, with CRT-shader and video capture. In Retroarch settings, I choose the target resolution, activate video lossless record + GPU shader record. As Retroarch won't play zmbv video, I first transcode to "lossless" h264, also setting the aspect ratio for the next pass. After Retroarch, I transcode the video to h264 Q1 to reduce the video size from 30GB to 10 GB, with no visible quality loss. Here is the makefile I used for the conversion: https://gist.github.com/kassoulet/485ce8bb3c29461ae67a5aeb5a683fbe (remember, the audio is taken from a video I generated using the remastering done in part one)...
Read more

Python bdist_rpm & changing the package name

When you create a RPM package with python's distutils, there is something rater annoying: you can't change the package name. or at least not easily. After a moment of bad/hard solutions, I've found a easy and clean way to do it: Let's say you have a customized nose. When you create the RPM with "python setup.py bdist_rpm", the resulting package is nose-$version.rpm, but you want it to be python-nose-$version.rpm, so it can be recognized by redhat packages. But a src.rpm is also created! You just have to install (rpm -i) the src.rpm, this will create a .spec file in ~/rpmbuild/SPECS and a tar.gz in ~/rpmbuild/SOURCES. Now, edit the package name in the spec file, decompress/rename the folder/recompress the tar.gz. Run rpmbuild -ba ~/rpmbuild/SPECS/yourspecfile.spec and the new/renamed RPM is in ~/rpmbuild/RPMS.
Read more

Stars Audio Remastering

Image
I present to you an experiment I did to create the best possible video for an 1995 MSDOS demo. Note the choices I made: Use Dosbox Process the audio from the original modules. Apply CRT emulation to the video. This post will detail the audio processing. Audio Source Choice The original demo uses two Amiga modules (MOD) for the sound, so there is no real panning. Some channels go to the left output, the others to the right. The demo is played in stereo on GUS soundcards, and in mono on SoundBlaster soundcards. For me the panning is questionnable, and after a few tests in mono, I go for a reduced stereo field + mono low-end. For the source I will use a high-quality replayer  with 8-tap fir interpolation to re-generate from the source modules. I also go for a full remastering of the track, as we are taking about one of the finest  masterpiece of French heritage here. (yeah I know the musician is not french :) ) Audio Edit The first step is to merge the two tracks into one, and ch...
Read more