Taperssection.com
Gear / Technical Help => Post-Processing, Computer / Streaming / Internet Devices & Related Activity => Topic started by: if_then_else on February 13, 2025, 12:47:18 AM
-
As suggested by "TheJez" I'm creating a new Thread for my TrackAlign tool which is based on the Python module Audalign.
I think I managed to fix some of the issues that have been addressed by some fellow Taperssection members in the VST plugin thread.
The current code and documentation can be found on
https://github.com/mmathis76/TrackAlign/
The updated *.exe file for Windows can be found here:
https://github.com/mmathis76/TrackAlign/releases
The manual is available here:
https://github.com/mmathis76/TrackAlign/blob/main/README.md
A few notes:
- As some people have already realised, the pydub module I have used to split and merge the stereo channels has an implicit dependency on ffmpeg or its fork avconv. I can't compile it into the exe file but might be able to create an NSIS installer for Windows that includes ffmpeg. For the moment you'll have to install it by yourself (unless it's installed already).
- I added a few new parameters that will allow you tp select one of the algorithms supported by Audalign (fingerprint, correlation, correlation_spectrogram, visual) and the accuracy level (default=3, higher=more precise)
- "fingerprint" can push your PC/laptop to very memory usage (depending on the length of the tracks and the number of tracks to align)
- "correlation" is a lot lighter in terms of memory usage and appears to yield similar results
- I noticed some "errors" with Audalign when using "correlation_spectrogram". However these had no noticeable impact on the resulting files
- As with any other tools, the results / accuracy may vary greatly with the input files. I tried the tool with 3 of my matrix recordings. With one of the samples provided by theJez there was some inexplicable glitch/inaccuracy on one track when using the "fingerprint" or "correlation" algorithms. This didn't happen with "correlation_spectrogram", though.
-
following
-
Hi if_then_else,
Thanks for updating the program. Just to let you know: I've tested the new version with the correlation_spectrogram method on my audience/mixing desk recording and unfortunately got similar results as with the fingerprint method: An offset of ~100-200ms between the two recordings. I did not use the '-c' option.
When I have the chance I will try with -c and the correlation method...
-
There's no guarantee that track alignment will work with every input file. As mentioned I encountered similar issues with yours but it worked fine on every other wav file I had tried.
That said, the maintainer of Audalign replied to me on Github re. a fix to his own module that would render stereo output files. This might not resolve the issue with your specific file. However, it might eliminate the need for file splitting and merging and the associated overhead. I'm in Belgium this week. Will have a look at the proposed fix when I'm back.
-
There's no guarantee that track alignment will work with every input file.
That I understand! I'm not expecting any guarantee and as I told you earlier I've manually alligned this specific recording years ago. I was just curious about the program and this MX/Audience recording was one of the few matrix recordings that I had available. I may try another matrix of two audience recordings of another show that I have. Again, just out of curiousity :-)
Thanks for the hard work on this program! It certainly makes life easier if it happens to work well on the input files at hand... :cheers:
-
Thank you, will check it out during the weekend.