Listening & research / Guide
Budget an AI-music workshop around critique, not output volume
A small-group workshop needs a credit ceiling, authorship notes and time to listen back—not a race to create the most files.
3 min read 19 Sep 2026 draftListening & research
Listen closely. Read critically.
Separate the attractive demo from the repeatable result. Explore artist experiments, evaluate model claims and design listening comparisons that answer a real question.
Listening & research / Guide
Benchmark stem separation fairly: score the signal and hear the job
A fair comparison names the test set, prevents leakage, reports the metric and then listens for the artefacts that metric may miss.
3 min read 19 Sep 2026 draft
Listening & research / Guide
Restore audio without rewriting its history
Preservation work needs an original, a documented derivative and a trail that explains every transfer, repair and listening copy.
3 min read 19 Sep 2026 draft
Listening & research / Guide
Read an AI-music study without mistaking a result for a verdict
A paper's sample, genre, task, metrics and listener controls tell you what it found—and what it cannot establish about music in general.
3 min read 19 Sep 2026 draft
History / Archive
A 1983 protocol let rival instrument makers speak one language
The MIDI Association's own specification pages date and describe the shared message format that made instruments from competing makers interoperable.
3 min read 1983 record · prepared 16 September 2026History / Archive
An oil-industry pitch algorithm was patented as a vocal corrector
US patent 5,973,252 documents the pitch-correction method behind Auto-Tune, traced by its maker to geophysical signal analysis.
3 min read 27 October 1997 event · prepared 16 September 2026Platforms & listening / Archive
YouTube's own blog places Content ID's launch in 2007
A YouTube retrospective and current help pages date the fingerprint system's start and describe what a match can and cannot decide.
3 min read 2007 record · prepared 16 September 2026Platforms & listening / Archive
Discover Weekly paired personal taste with similar fans' plays
Spotify's 2015 launch announcement and its follow-up usage data describe how the weekly playlist blends personal and collective listening signals.
3 min read 20 July 2015 event · prepared 16 September 2026Models & research / Archive
NSynth shipped as an open dataset and model, not just a demo
Google's Magenta team published the NSynth paper, dataset and code together, letting musicians inspect and rebuild the instrument.
3 min read 6 April 2017 event · prepared 16 September 2026Models & research / Archive
WaveNet generated raw audio, and then music, sample by sample
DeepMind's paper and blog post describe a model predicting each audio sample from prior ones, applied first to speech and then to music.
3 min read 8 September 2016 event · prepared 16 September 2026Artists & releases / Archive
Daddy's Car credits an AI draft and a human arranger separately
Sony CSL's own archive records that Flow Machines generated the composition and a named musician arranged and produced the final track.
3 min read September 2016 event · prepared 16 September 2026Models & research / Archive
Jukebox generated singing from raw audio and stayed a demo
OpenAI's two music systems show a research trajectory from MIDI notes to raw audio, released with unusually candid limitations.
3 min read 30 April 2020 event · prepared 16 September 2026Models & research / Archive
A hobby fine-tune turned an image model into a music one
Riffusion repurposed Stable Diffusion to paint spectrograms, showing image diffusion could generate usable audio.
3 min read 15 December 2022 event · prepared 16 September 2026Models & research / Archive
MusicLM was published with its memorisation study attached
Google's text-to-music paper reported a duplication test and withheld the model for over three months.
3 min read 26 January 2023 event · prepared 16 September 2026Models & research / Archive
Meta open-sourced MusicGen with a licensed dataset and an NC licence
AudioCraft released MusicGen's weights and code but split their licences, and named its training data source.
3 min read 2 August 2023 event · prepared 16 September 2026Models & research / Archive
Stable Audio switched from a licensed library to Creative Commons
Stability AI's two audio models used different training-data bases and different commercial-use ceilings.
3 min read 13 September 2023 event · prepared 16 September 2026Platforms & listening / Archive
Dream Track built a generator around named, consenting artists
YouTube and DeepMind's Lyria experiment paired opt-in artist consent with SynthID watermarking on every clip.
3 min read 16 November 2023 event · prepared 16 September 2026Models & research / Archive
Open source made stem separation common but not a rights grant
Spleeter and Demucs put studio-grade separation on any laptop, with Spleeter's own docs warning it is not a licence.
3 min read 28 October 2019 event · prepared 16 September 2026Artists & releases / Archive
Holly Herndon released her own voice as a governed instrument
Holly+, launched in 2021, turns audio into Holly Herndon's voice under a governance model her own studio describes but does not itemise.
3 min read July 2021 event · prepared 16 September 2026Artists & releases / Archive
Now and Then used separation, not generation, to finish a demo
Apple Corps says machine-learning source separation freed John Lennon's 1970s vocal from his piano so The Beatles could finish the song.
3 min read 2 November 2023 event · prepared 16 September 2026Platforms & listening / Archive
YouTube extended Content ID-style detection to AI likeness
YouTube opened likeness detection to Partner Program creators in September 2025, modelled on Content ID but limited to detection.
3 min read 16 September 2025 event · prepared 16 September 2026Platforms & listening / Archive
Spotify pulled Boomy tracks over suspected stream manipulation
Trade reporting and Spotify's own policy show a 2023 Boomy takedown targeted artificial streaming, not the fact that Boomy's music was AI-generated.
3 min read 1 May 2023 event · prepared 16 September 2026Platforms & listening / Archive
Spotify set new rules on AI voice clones, spam and disclosure
Spotify's 25 September 2025 announcement documents impersonation, spam-filter and AI-credit policies, without a ban on AI-made music itself.
3 min read 25 September 2025 event · prepared 16 September 2026
Platforms & listening / Archive
Deezer began tagging albums that are fully AI-generated
Deezer's own announcements document a detection tool, a June 2025 tagging system, and the rising delivery share it says the tool measures.
3 min read 20 June 2025 event · prepared 16 September 2026Platforms & listening / Archive
A DDEX-built credit standard is how Spotify's AI labels arrive
Spotify's own announcement documents an AI-disclosure credit standard developed through DDEX, delivered by distributors rather than typed by a listener.
3 min read 25 September 2025 event · prepared 16 September 2026Artists & releases / Archive
The Velvet Sundown went viral before its AI origin was confirmed
Billboard's reporting documents how an AI-generated band gained streams and press attention weeks before a disputed on-record confirmation.
3 min read June 2025 event · prepared 16 September 2026Platforms & listening / Archive
IFPI recorded a ninth year of recorded-music revenue growth
IFPI's 2024 report documents 10.2% revenue growth in 2023 and names generative AI as a risk fans and the industry want managed responsibly.
3 min read 21 March 2024 event · prepared 16 September 2026Platforms & listening / Archive
Spotify publishes Loud and Clear as its own royalty evidence
Spotify's Loud and Clear site documents how it defines payouts, the context it supplies, and what a maker should verify before citing its numbers.
3 min read Practice record · prepared 16 September 2026Platforms & listening / Archive
Spotify stopped paying royalties on tracks under 1,000 streams
Spotify's own documents show a track needs 1,000 annual streams and a minimum listener count to earn recorded royalties, effective since April 2024.
3 min read 21 November 2023 event · prepared 16 September 2026Platforms & listening / Archive
YouTube required creators to disclose realistic synthetic media
YouTube's blog and help centre document a disclosure requirement for realistic AI content, explicitly listing AI-generated music among it.
3 min read 18 March 2024 event · prepared 16 September 2026Platforms & listening / Archive
Bandcamp barred substantially AI-generated music from its store
Bandcamp's terms bar training on artists' music without permission, and reported January 2026 statements bar selling substantially AI-made music.
3 min read 14 January 2026 event · prepared 16 September 2026
Artists & releases / Archive
YACHT turned three years of machine-generated fragments into an album
The band's own DFA notes and Bandcamp page describe building an AI process, then choosing which fragments became songs.
3 min read 30 August 2019 event · prepared 16 September 2026Artists & releases / Archive
Eight songs on one album named four different AI tools
The artist's own site credits Amper, Watson Beat, AIVA and NSynth track by track for the 2018 album I AM AI.
3 min read September 2018 event · prepared 16 September 2026History / Archive
A composer coded recombination decades before neural models
Cope's own archived pages describe starting Experiments in Musical Intelligence in 1981 and its successor, Emily Howell.
3 min read 1981 record · prepared 16 September 2026History / Archive
An orchestra recorded a score its computer wrote from scratch
Melomics' own pages date the 2011 chamber premiere and the 2012 London Symphony Orchestra recording of computer-composed scores.
3 min read 13 April 2012 event · prepared 16 September 2026
Artists & releases / Archive
PROTO trained its AI voice on consenting singers in the room
The artist's Bandcamp notes and studio page describe training the Spawn model live with volunteer singers rather than existing recordings.
3 min read 10 May 2019 event · prepared 16 September 2026History / Archive
Eno set rules running in 1996 and called the result generative music
A 1996 talk transcript and Eno's own 2017 app notes define generative music as rules set in motion, not a fixed recording.
3 min read 8 June 1996 event · prepared 16 September 2026Artists & releases / Archive
A mental-health charity credited AI, not estates, for four songs
The campaign's own FAQ says an AI algorithm learned from 27 Club musicians' catalogues, while disclaiming estate endorsement.
3 min read April 2021 event · prepared 16 September 2026Artists & releases / Archive
A settlement turned Suno into Warner's licensed AI platform
WMG and Suno's own statements describe a litigation settlement and an opt-in framework, without yet naming a participating artist.
3 min read 25 November 2025 event · prepared 16 September 2026Platforms & listening / Archive
Fingerprint matching finds registered recordings, not resemblance
Audible Magic and Content ID documentation describe matching acoustic fingerprints, not judging style or originality.
3 min read Practice record · prepared 16 September 2026Models & research / Archive
Two decades of shared MIR evaluation still exclude generation
MIREX's own 2026 call for proposals says generative-model evaluation is a gap, not a solved comparison.
3 min read Practice record · prepared 16 September 2026Models & research / Archive
A shared dataset made separation scores comparable, not perceptual
MUSDB18 and the SDR metric let separation systems be ranked by decibels, which the campaign's own paper says diverges from listening preference.
3 min read 17 December 2017 event · prepared 16 September 2026Models & research / Archive
A generation metric adapted from images admits a moderate correlation
The Frechet Audio Distance paper reports a 0.52 correlation with human ratings, calling it better than SDR, not a proxy for listening.
3 min read 17 January 2019 event · prepared 16 September 2026Models & research / Archive
A neural codec's tokens explain a generator's sonic signature
EnCodec compresses audio into discrete codes that generators like MusicGen predict, so the codec's decoder shapes the final sound.
3 min read 24 October 2022 event · prepared 16 September 2026Platforms & listening / Archive
Streaming loudness targets sit on top of an ITU measurement standard
Spotify's -14 LUFS policy uses the ITU's BS.1770 algorithm, but the standard sets no target level itself.
3 min read Practice record · prepared 16 September 2026Platforms & listening / Archive
An open database made music credits editable, not automatically true
MusicBrainz stores CC0-licensed, volunteer-edited relationships, which helps disputed credits but is not a verified registry.
3 min read 17 July 2000 event · prepared 16 September 2026Platforms & listening / Archive
Epidemic Sound owns its catalogue instead of collecting royalties
The company's own pages describe a direct-licensing subscription model that clears rights before a track is ever streamed.
3 min read Practice record · prepared 16 September 2026Platforms & listening / Archive
Streaming brought in 84% of US recorded music revenue in 2023
RIAA's own year-end statistics show streaming's share, the paid-subscription split, and what physical formats still earn.
3 min read 26 March 2024 event · prepared 16 September 2026Platforms & listening / Archive
Deezer and UMG stopped paying noise the same as songs
The companies' joint announcement redefines which streams count as royalty-worthy, and by how much.
3 min read 6 September 2023 event · prepared 16 September 2026Platforms & listening / Archive
TuneCore's terms name which AI platforms it will not distribute
The distributor's own Terms of Service set a training-data-licensing test that decides whether generated music can go to stores.
3 min read Practice record · prepared 16 September 2026Models & research / Archive
The Copyright Office found prompts do not control output
The Office’s Part 2 report and two generators’ own documentation show why a prompt describes an idea but does not direct its expression.
3 min read Practice record — prepared 16 September 2026Platforms & listening / Archive
Music’s AI credit standard is still being built by DDEX
DDEX confirms a metadata standard for AI disclosure is still in development, while Spotify and YouTube already run their own separate labels.
3 min read Practice record — prepared 16 September 2026Models & research / Archive
A broadcast standard specifies how to blind-test audio
ITU-R’s MUSHRA recommendation gives a small studio a tested method for comparing generated and human tracks without fooling itself.
3 min read Practice record — prepared 16 September 2026Artists & releases / Archive
The Grammys ruled an AI voice cannot itself win an award
The Recording Academy's own interview and current rulebook require meaningful human authorship before AI-assisted work can win.
3 min read 17 June 2023 event · prepared 16 September 2026Platforms & listening / Archive
DistroKid allows AI music but requires a disclosure credit
The distributor's own help centre states what counts as AI, what must be disclosed, and what still gets a release rejected.
3 min read Practice record · prepared 16 September 2026