The Conservatory

Upload a track

Hear what a producer hears.

Drop a track you hold. You get the tempo, the key, the chord loop and how it ends, the groove and the mix — written the way a producer would say it. Every value is an estimate from audio; tempo and key carry their confidence, and when the ear can’t decide it says so.

or drag it here

WAV, MP3, M4A, FLAC, OGG, AIFF or AAC · up to 40 MB · the first 90 seconds are read · your file is deleted when the read is done.

Want the parts — and a new track in this DNA?

The deep ear splits your track into stems (drums · bass · other · vocals) and gives you MIDI for each — pitched stems transcribed note by note, the kit as GM drum hits from its onsets — importable in FL Studio, Ableton or Logic. Tell it what to change — “10 bpm faster and more orchestral”, “darker, in D minor, swing it” — and it composes a new track in the same key, tempo, loop and energy, bent to your instruction, then listens back to audit itself. Leave an email; it usually builds in a few minutes and downloads as a ZIP.

Choose a track above first.

Not done, on purpose: vocals, “sound like artist”, copying or extending a recording. What you get back is your new work.

What it reads

  • Tempo and how steady the pulse is
  • Key, with the runner-up when it’s close
  • The repeating chord loop as Roman numerals — and the cadence it ends on (authentic, half, plagal, deceptive)
  • Chord qualities: triads and sevenths
  • Groove cell, backbeat and swing — estimated from the mix; read from the drums stem in the deep tier
  • Sections and the energy arc
  • Mix: integrated LUFS, stereo width, brightness, low-end centre, dynamic range

How it’s built

  • Every value is an estimate from audio and says so; when it can’t decide, it says that instead of guessing
  • The chord loop is checked by the same theory verifier that gates the composer — the ear and the maker speak one language
  • Engines are license-clean: Demucs (MIT), basic-pitch (Apache-2.0), librosa, music21, plus the first-party Bonis Systems composer, groove, mix and BS.1770 mastering code
  • Your audio never leaves this service — no third-party AI API sees it. Attributions at /licenses

Questions

Do you keep my audio?

No. Your upload is processed in a temporary folder and deleted when the request finishes. The Conservatory keeps the read you were shown and — for the handback only — your email, your instruction, and the requesting IP address. Data questions: privacy@bonissystems.com.

What does the producer’s read include?

Tempo and how steady it is, the key with a runner-up, the repeating chord loop written as Roman numerals and the cadence it ends on, chord qualities (sevenths etc.), the drum groove cell and swing, sections, and mix metrics: integrated LUFS, stereo width, brightness, low-end centre and dynamic spread.

Is it free?

The read is free with no account. The handback (stems, MIDI per stem, optional collab track) needs an email and downloads as a ZIP.

How accurate is it?

Every value is an estimate from audio and is labelled as one; the read shows a confidence for tempo and key and names its caveats. Chords are the best template match on each bar’s harmonic content; the loop is the shortest cycle that mostly repeats. In testing on produced tracks with a clear pulse, tempo, key and loop reads matched the source; on rubato or dense material it will tell you it’s unsure.

Can it make a track that sounds like a specific artist?

No, on purpose. It composes a new track from the heard DNA (key, tempo, loop, groove, energy) plus your instruction, through license-clean engines, with every progression theory-verified. It does not copy, extend or imitate a recording, and it does not do vocals.

What is the collab track, exactly?

A new composition: the heard key, tempo, loop, groove and energy arc become a brief; your instruction bends it; a theory-verified composer writes harmony, a shaped melody, bass and a kit; real sampled instruments render the pitched parts (the kit is synthesized); it’s mastered to −16 LUFS and the achieved figure is written in the READ; the ear listens back to check tempo, key and loop against the brief; and you get a verifiable record link. It’s yours.

Can I run this on my own machine?

Yes — it’s a Python engine. Write to contact@bonissystems.com.