[Feature]: Pitch / key transposition for vocal practice and karaoke

[thcp]

What problem does this solve?

There's no way to change the musical key of a track. This is essential for vocal practice and karaoke:

• Singers need to lower songs with high notes to practice safely before attempting the original key.
• Users want to transpose backing tracks to fit their natural vocal range when it doesn't match the original artist's.

Pairs naturally with the existing stem separation + speed control: separate stems → mute vocals → transpose the backing track to your range → practice at your tempo. Key change is the missing piece that turns the stem separator into a real vocal-practice tool.

Proposed solution

Add a pitch / key transposition control (e.g. a ±6 or ±12 semitone stepper with a key label) alongside the existing playback speed control — pitch shift without changing tempo, the complement to the speed control (tempo without pitch) shipped in #241 / #242.

Implementation notes:

• The audio engine already runs a WSOLA time-stretcher in an AudioWorklet (static/vendor/soundtouch-processor.js), but it implements the tempo stage only — no resampler / pitch parameter. Pitch shift needs a resampling stage added (the standard SoundTouch time-stretch + resample design).
• Recommended: add a pitch parameter + interpolating resampler inside the worklet, keeping pitch out of the delicate playback-rate/scheduling math in the audio engines.
• Start real-time, client-side, scoped to backing-track transposition — instrumental artifacts from WSOLA + resampling are far less objectionable than on isolated vocals. Higher-quality, formant-aware vocal shifting can be a later server-side enhancement.

Alternatives considered

• Prior art: the Transpose browser extension, and Moises.ai's key-change feature.
• Compose existing primitives (playbackRate = 2^(N/12) on the source + SoundTouch tempo = 1/2^(N/12) to restore duration) instead of adding a resampler — quicker, but tangles pitch into the existing scheduling math, which is a likely source of bugs.
• Server-side offline render for maximum quality (formant preservation) — heavier and slower; better as a follow-up if users want to transpose isolated vocals well.

[gerrybrano]

To add to "What problem does this solve?", studio recordings are sometimes sped up or slowed down to get the tempo "in the pocket" or to add a "sparkle" to the mix. For example, "Ob-La Di, Ob-La Da" by The Beatles was performed in the key of A but the tape was sped up and sounds like A#, which is a harder key to perform it in.

It is also worth noting that this technique would often yield recordings that to not align with A440 pitch standard, such as "Another One Bites the Dust" by Queen, so a control that is quantized to semitones would not help in this case.

[joggs]

Also, when changing playback speed, the key should be kept, so you don't get that smurf effect E.G increasing the pitch with the same amount as you increase the speed. Perfect while training fast guitar solos for example.

[thcp]

Hey @joggs / @gerrybrano

not forgotten, just still on vacations here :)

[gerrybrano]

@thcp Dude, you owe us nothing. You've provided a really amazing contribution to music studies and asked for nothing in return. You may not think it's a big deal, but you've likely made a bigger impact on people's lives than you realize. Thank you truly. Now... what are you doing here?! Enjoy your vacation!

[thcp]

@gerrybrano really kind of you to write that, that's exactly the reason why I'm committed to keep stemdeck free !

[adlez27]

I'd like to note that the drum stem doesn't necessarily need to be transposed when changing key, just everything else. Often the reason I choose to split a song into multiple stems instead of just vocal/instrumental is to avoid distorting the drums when I want to change the key.