Moises: stem separation built for practising musicians

Moises does one thing that used to require the original studio session: it takes a finished recording and pulls the parts back out of it. Feed it any track and you get the vocal on its own, the bass on its own, the drums on their own. Behind it sits a neural network retrained on larger licensed datasets, and the newest version detects more than 30 instrument and vocal categories rather than the three the app started with.

The company behind it is Music AI, founded by a Brazilian team led by Geraldo Ramos and shipping since 2020. Its 2025 recap puts the user base at 70 million, with 15 million joining that year, 1.2 billion minutes of audio processed, $50m raised, and availability in 33 languages across 190 countries. The top five markets are the United States, Brazil, India, Mexico and the UK. Charlie Puth joined as chief music officer, Ableton built the separation into Live 12.3, and in October the company announced a tie-up with SoundCloud.

Built for the player, not the producer

The aim was never the producer. It was the person practising.

“There are two futures for AI in music. In one, you type a prompt and a machine gives you a song. In the other, AI helps empower musicians to play, experiment, and create together while keeping humans at the center.”

— Geraldo Ramos, co-founder and CEO of Music AI, https://moises.ai/newsroom/company-milestones/2025-year-in-review/

That distinction is the whole product. If you play bass in a covers band, the useful thing is not a generated song; it is tonight's set list with the bass stripped out so you can play against the rest. Around that sit a metronome, key and tempo changes that do not turn the recording into a chipmunk, chord detection, lyric transcription and a video capture mode.

So it is for people who play: the guitarist learning a part by ear, the singer who needs a backing track without the original vocal, the teacher building exercises, the DJ after a clean acapella, the editor lifting speech out of a noisy recording. It is not for someone who wants a machine to write the song.

What it costs and where it breaks

Price is the other reason it spreads. Premium is $3.99 a month and Pro, which adds top-quality separation and API access, is $9.99. We checked that against the rest of our own AI catalogue: of the 40 services there with a monthly subscription, Moises has the cheapest paid tier of all of them, and the free tier takes five uploads a month without asking for a card. The ceiling is the same as the floor, though. Separation quality falls apart on dense mixes and on instruments that share a frequency range, and no model reconstructs what the mixdown destroyed. For rehearsal that rarely matters. For release-grade work it still does.

Informational material, not investment advice and not a paid placement. Prices were checked on 24 October 2026, and the user and processing figures are the company's own published numbers.