Tero Parviainen

One half of @ctpt_studio. Building music systems for wellbeing and movement.

[Tero Parviainen @teropa]

We’re announcing something new today: Motif - music that moves with your yoga class.

Instead of static playlists, the music follows the structure and energy of a class and adapts on the fly.

[Tero Parviainen @teropa]

@crazyjoedavolax i needed a good timestretch in my music system. i made one. what's the problem?

[Tero Parviainen @teropa]

Made a rather serviceable timestretching algo for drum loops with Fable, some web research, and a few rounds of listening.

Cut the loop at every hit, keep the hits sample-for-sample, and stretch only the decays between them. Separating noise and ring.

[ʏ 𝗔 ⊂ ᴴ ᴛ̞ @YACHT]

📡🌴 NEXT WEEK on Friday the 13th we're throwing a party to celebrate the vinyl release of Greatest Hits from Planet Blip and you're invited! If you're in, near, or willing to beam yourself to Los Angeles, we would love to see you.

FREE & all-ages! RSVP:
https://t.co/dohBLYCfWb

[Tero Parviainen @teropa]

@crazyjoedavolax lol i'm perfectly aware of the prior art

[Tero Parviainen @teropa]

@ButeSound Fair enough. I would posit though that messing with different algorithms applied to different material can also be a path to creative discoveries.

Infinite roads to infinite destinations really.

[Tero Parviainen @teropa]

@ButeSound ruined what loops and for whom?

[Tero Parviainen @teropa]

@theycallmeken no i wouldn't claim that. it's a solution to a problem i had, made from existing parts.

[Swann Bessa @swann_bessa]

Introducing e-means, an algorithm beating k-means in 10x fewer epochs.

With order of magnitude wall-clock speedups, e-means is released in FAISS and needs no hyperparameter tuning beyond k. It is robust to various scales and datasets. Still skeptical ? Just try e-means [1/10]🧵

[Tero Parviainen @teropa]

@jakepenn_ thx! runs about 11x real time for me here. offline use so it's plenty fast enough.

[the_simian @5imian]

We had better AI music 40 years ago. Beyond the musical/style errors others have noted, this is only impressive if you don't know the field. A Markov chain trained on Bach does better. That was our LISP "hello world" when I studied with David Cope 20 years ago
It really does suck

[Tero Parviainen @teropa]

@thepatch_kev yeah it does. i'm only using it for drums, and using rubberband-r3 warps for other stuff

[Tero Parviainen @teropa]

@thepatch_kev thx yeah performed well across our material, happily deploying it now

[Tero Parviainen @teropa]

@FrankTweets2026 yeh the slicing part of this goes all the way back

[Tero Parviainen @teropa]

@facesOfEth inverse is mildly cursed

[Tero Parviainen @teropa]

it's embedded in a bigger system, alas no standalone thing to download. but claude can probs do it from "Cut the loop at every hit and keep each hit sample-for-sample, so nothing smears. Only the decays between hits get stretched, in two parts: the noisy part (snare rattle, hats, room) by overlapping short grains, and the ringing part (kick body, toms, bells) by a phase vocoder that picks up in phase with the hit before it. Slicing is what DAW beat modes do and the noise/ring split is from Driedger, Müller & Ewert (2014); the combination is what made the tails clean."

[Tero Parviainen @teropa]

@loamyclay it's a good method, try and break things in interesting ways

[Tero Parviainen @teropa]

@tenshi_exe how very astute of you

[Tero Parviainen @teropa]

@alexanderchen Yeah, I love that you can now build some intuition without needing deep paper studies every time. Changes the opportunity cost economics.

[Tero Parviainen @teropa]

@akx 0 -> 0000000

sorted ✅

[Tero Parviainen @teropa]

@reallymeals thx! this "tufte skill" basically one-shotted the viz https://t.co/h3y1fCXlQ5

[Tero Parviainen @teropa]

more tonal material in a tabla loop

[Tero Parviainen @teropa]

Not invented from thin air: slicing is what DAW beat modes do. The noise/ring split is from Driedger, Müller & Ewert (2014).