[hasezoey]

This PR adds wsola experimentally as a pure rust tempo stretcher alternative to rodio's plain convert and FFI soundtouch.
Additionally, cleans-up the sources a bit.

Unlike #793, in this PR everything works, but has abyssal performance in debug builds, either requiring upstream changes, or using opt-level for this specific dependency.
Additionally, in its current state, it is missing the 1.0 (default) tempo bypass, which would also improve performance.

fixes #167
closes #793

This is a DRAFT for testing and to hopefully fix the abyssal debug build performance (without cargo opt level).

[tramhao]

You translate the algorithm yourself? Excellent!

[hasezoey]

You translate the algorithm yourself? Excellent!

I did not. Since about 3 months, there exist 2 crates which had implemented the algorithm, both released about the same time. I just came across them and pushed some PRs to improve them.

PS: i converted it back to a DRAFT as the debug build performance is still not solved.

[hasezoey]

I think wsola(crate) itself is now ready after all the changes. I updated the PR to the latest version.
It is ready for a test listen again, but before merge i want to add a bypass for normal speed (like soundtouch has), so we dont unnecessarily use CPU or modify the samples.

While working on this and the other PR, i noticed how bad the actual rodio naive speed change actually is.
@tramhao do you want to make this the default and remove the rodio naive speed change option, or keep it? (likely done in another PR)
Note that this is not about the soundtouch option, which i would like to keep as it does sound slightly better than wsola.

[hasezoey]

Likely final update:
Implemented a bypass, like in soundtouch, which only bypasses until the algo is started.
Due to how wsola works, at least to my understanding, there is a slight audio distortion at algo start (ie once you first change the speed to non-1.0), but it is so slight that i think it is acceptable. (please test)

Additional changes:

• always optimize wsola in debug mode, as otherwise it consumes way too much CPU while running
• update soundtouch impl to not accidentally change to algo in the middle of a frame
• cleaned-up git commits
• Explicitly log what Timestretcher is active, not just when "soundtouch" is active

For caution, this will stay behind the TERMUSIC_RUSTY_WSOLA environment variable flag, until making it the default in a follow-up PR, as said in #794 (comment).
(Set TERMUSIC_RUSTY_WSOLA=1 to enable WSola, but only if soundtouch is either not compiled in or set to active in the config)