I Built a Local AI Music Studio

Description and use

Miso is a local AI music studio that integrates multiple models and post‑processing steps into a single workflow, enabling users to generate, edit, and remix music without command‑line tools.

Problem

Users want an interface to experiment with local music models and manage post‑generation tasks such as editing, stem separation, voice conversion, and MIDI transcription without juggling command‑line tools.

Outcome

Miso enables a streamlined local‑first workflow for AI music creation, editing, and remixing, integrating multiple models and post‑processing steps into a single interface.

Workflow

• 1. Create a take by prompting a model (e.g., ACE‑Step 1.5).
• 2. Use guided builder to translate UI controls into model‑specific syntax.
• 3. Edit the generated audio (trim, fade, normalize) directly in the browser.
• 4. Repaint sections by selecting waveform ranges and regenerating only those parts.
• 5. Split the track into stems (vocals, drums, bass, etc.).
• 6. Transcribe a vocal stem to MIDI notes.
• 7. Convert MIDI to an ABC score and inspect tempo/key.
• 8. Feed the score to YuE2 for a new arrangement.
• 9. Mix stems, apply voice conversion, and export the final take.

Evidence

Miso is a self‑hosted local music generation and remix studio built on top of audio.cpp.

Source

The whole thing runs as two Docker containers. One runs Miso itself. The other runs audio.cpp and owns the GPU models.

Source

Miso can also split a track into stems.

Source

The current release, 1.4.0, connects MIDI transcription to YuE2's score support.

Source

I generated a simple version of *Twinkle Twinkle Little Star*, ran it back through separation, transcription, and score conversion, and recovered the expected pitches across the recording.

Source