Work Web app · Audio & AI

Practice Studio: vocals & drums

Helping a singer and a drummer practise any song exactly the way they need. Two practice apps on one engine: paste a song, AI splits it into vocals, drums and the rest, then you rehearse with pitch, tempo, loops, a metronome and synced lyrics. The singer also writes her own sheet music in it.

Role
Idea, design and development (built for people close to me)
Year
2026
Status
Personal project
Type
Web app · Audio & AI
A metronome built into the app
The sheet music editor on a phone
Song editor: waveform with sections, pitch, tempo, mixer and synced lyrics
Sheet music editor: voice and piano, lyrics under the notes and an on-screen keyboard (demo score)
View more screens (3)
Full-screen karaoke mode with the current line highlighted
The drum version: separate drums, vocals and the rest, plus a metronome track
Song library with detected key and BPM

The challenge

A good backing track in the right key, or a version of a song without the drums, usually means hunting through YouTube or paying per song. Practising a tricky passage means scrubbing back and forth by hand.

What I built

A web app where any song becomes a practice session. The heavy audio work happens in the cloud, so the app opens instantly on an iPhone, iPad or laptop.

Outcome

In regular use by the singer and the drummer it was built for, each with their own version of the app on the same architecture.

Key features

  • Separate the vocals, drums and other instruments of any song
  • Change the key or speed without starting over
  • Repeat difficult sections until you get them right, with markers for verses and choruses
  • Drum mode: mute or solo the drums, vocals or the rest, with a metronome
  • Lyrics generated from the song, a karaoke mode and audio export
  • Write, play back and export your own sheet music
  • Songs are prepared in the background, so the app stays quick to use

Built with

  • React
  • TypeScript
  • Hono
  • Cloudflare Workers
  • D1
  • R2
  • Python
  • Demucs
  • Whisper
  • Modal (GPU)
  • abcjs
Technical details
  • AI stem separation with Demucs; key and BPM detection; lyrics transcription with Whisper
  • Pitch shift of ±12 semitones and independent tempo control; A-B loops on the waveform
  • Audio processing on a cloud GPU (Modal), with an automatic fallback worker
  • Sheet music drawn and played with abcjs; export to PDF, PNG and MIDI

Next project

Momentum Church →

Have a project in mind?

Tell me a little about it. I’ll get back to you within one working day. If we’re a good fit, we’ll discuss the details and I’ll prepare a clear, written quote.

Start a project →[email protected]