dbry / audio-stretch Goto Github PK

Looking at the commit in which you extended the stretch range from [0.5, 2.0] to [0.25, 4.0], I'm understanding that the limits don't come from just what's considered reasonable quality, but that the implementation depends on those limits, and changing them requires a reworking of the source code.

Am I right, or the current implementation could allow for extending the stretch for longer than 4x without changing the source?

My interest, as said in the title, would be for videos in super slow motion. Agreed, at longer than 4x, the audio you can expect is going to be quite weird, but I'd like to have it. Or maybe do you recommend another time-stretch code/algorithm for doing that?

I have no interest in stretching below 0.25x, because I'm talking about slow motion only. So, it's the 4x limit the only one I'd like to remove.

stumbled over the chrome plugin skip-silence, see https://github.com/vantezzen/skip-silence , which is a very interesting idea.
it allows to configure different speeds for (non) silent sections .. and a threshold for the silence detection.

a good (automatic) silence detection might get complicated. the manual threshold should be simple.
for futher simplicity, a preprocessing module might zero the "silent" sections, that the already existing zero-detector can be used ..

direct support for ratio up to 4.0 would be great - correspondig to speed x 0.25,
this is the range, which is offered by many players.

usually, there's just one single signal source - even when recorded stereo in 2 channels.
thus i assume, it's ok to sum both channels and search for the pitch afterwards. that does implicitly save cpu power.

in case, the channels are completely independent and might speak at same time (sort of crosstalk) .. i would assume, the stretching quality is degraded, right?
if this is the case, it would be beneficial to have an option to process (and buffer the results) independently.

I made a Python wrapper for audio-stretch: https://github.com/twardoch/audiostretchy and https://pypi.org/project/audiostretchy/

what would be useful unittests?
any simply generatable test signals?