Postdoc @MIT_CSAIL / prev. PhD @northwesterncs Interactive Audio Lab / audio & ml
Joined October 2021
- Tweets78
- Following200
- Followers183
- Likes396
always heartwarming to see someone discover the Mick Gordon Doom soundtrack making-of video
サイン波ひとつから、どこまで音を変化させられる？🧪
今回はセンドのフィードバックループを使って、たった1つのサイン波から実験的なサウンドを作ります。
⚠️ただし、この実験は急激な音量上昇が起こる可能性があります。スピーカー、ヘッドフォン、耳を傷めないためにも、試す前に必ず以下を守ってください。
・全センド＆マスターにLimiterを入れる
・センド量は1dBずつゆっくり動かす
・スピーカー／ヘッドフォンは最小音量にする
準備ができたら、各センドにディストーション、ディレイ、リバーブ、Erosionなどを入れ、少しずつお互いにフィードバック。
ノイズやレゾナンス、位相の変化が複雑に絡み合い、たった1つのサイン波が予想もしないサウンドへ。つまみを少し動かすだけで、刻々と表情が変わっていきます🔥
they got Claude doing dev lemons bits
this absolutely blew me away. there's so much to talk about but:
- the singing is shaped from code. to make sure it was legible claude used a local speech -> text model to check the outputs and tweaked in a loop
- all band members actually play their instruments, frame by frame, as accurately as cartoonishly possible
- claude was literally exporting waveforms as pngs and looking at them as part of the mixing process
- there's a ton of easter eggs in the music video to find I'l l cover in another post
the fact these things always default to The 4 Chords is worse news for alignment than anything in the METR report
looking forward to mass audiophile psychosis but for sensing non-distortionary entropy-dependent text watermarks instead of gold plated cables
Anthropic says new Claude models will embed invisible watermarks in all generated text, everywhere Claude is offered.
The watermark is part of the text, it isn't metadata: "it will travel with the text when it's copied and pasted elsewhere, and may persist through some editing."
This starts with models launched on or after August 2, 2026, under an EU AI Act code Anthropic signed. Anthropic is still working on adding it to current models. The rollout is worldwide.
everyone you know is vibecoding generative loopers with a b&w color scheme
ORAM v0.3 is out ✿
New additions:
- loop zone variations powered by @stableaudio inpainting
- proper master recorder
- fade in / fade out
- undo / redo
- per-layer reverse playback
- nuke now actually nukes everything
- quieter, quicker, prettier FX palette
- smoother GUI
of course the cool new causal latent GAN paper is from the Musika author (+friends)
real ones will remember
arxiv.org/abs/2208.08706
feel like at least 50% of this is just letting guitars & vocals annihilate the upper midrange in ways that would terrify today's soothe-brained producers
for future discourse farming I would recommend producing from scratch and then passing through an old VAE
This Post is from an account that no longer exists. Learn more
image folks ahead of audio as always teapawt.org/SHL0MS/status/20542806…
Slides and code for our Interspeech tutorial are up now!
Had a blast presenting a tutorial on speech watermarking at #Interspeech2025 in Rotterdam!
If you’re in town and want to talk watermarking, definitely reach out! And while you’re here, I recommend the pinball museum for the big lebowski sensory overload chamber
Had a blast presenting a tutorial on speech watermarking at #Interspeech2025 in Rotterdam!
If you’re in town and want to talk watermarking, definitely reach out! And while you’re here, I recommend the pinball museum for the big lebowski sensory overload chamber
train on strongly metered rhyming lyrics —> model learns to associate phrases with similar metric & phonetic structure, even if semantically nonsensical
very Max Martin coded tbh
Suno + Veo 3 generate highly similar versions of popular songs purely based on *phonetically* similar gibberish lyrics?!?!
Presenting Bob’s Confetti: Phonetic Memorization Attacks in Music and Video Generation
🔊: jrohsc.github.io/music_attac…
📖: arxiv.org/abs/2507.17937
🧵1/n