Testing g#238 pitch/stretch algorithm (audio files included)

[ulmutul]

I just tested g#238 (featuring Soundtouch Parameters) and compared it to the pitch algorithm currently in HEAD.
I found the patched one far better (for my usecase) than the original one.

I've uploaded some test files (recorded straight from an actual player):
https://www.dropbox.com/sh/94p3ywt6kq8xhzi/AAD9BglkLX9qpaato2kmyZ3-a?dl=0

• "SP" in the filename means "Soundtouch Params" or in other words the patched version (g#238).
• The files named "*-stretched.mp3" are stretched to 70% of the tempo (they become ca. 40% longer). Great for practicing a guitar solo at a slower pace.
• The files named "*-pitched.mp3" are pitch-shifted one halftone up (so you can play with a standard tuned guitar to downtuned recordings).

When listening to "Mabon" you'll notice that the original pitch algorithm produce some kind of graininess in the sound, but the really ugly things happen at 0:07 and 0:16, when the sound suddenly wobbles between pitched and unpitched. In "Duende" this happens from the beginning and makes the whole intro unlistenable.

When using more extreme settings - like in the time stretch example - the original algorithm is slightly superior. You'll notice that in "Maiden...-stretched SP" all snare hits sound like a flam (a kind of double stroke).

For my taste the patch would be a big improvement in sound quality, but some people may have different opinions and/or usecases, so I think this should be discussed before

[jhMikeS]

Is there a big demand for transposing or re-tempoing music? Admittedly, from the preprocessor defines in the patch it you can see it wouldn't be much of a burden to have an menu option to optimize it for music or voice. I definitely didn't want to degrade HEAD's quality for its originally intended use, which was voice.