I Built a Local AI Music Studio

Описание и применение

Miso упрощает эксперименты с локальными музыкальными ИИ‑моделями, объединяя генерацию, редактирование, разделение стемов и транскрипцию в одном веб‑интерфейсе.

Задача

Users want an easy way to experiment with local music generation models and edit the results without juggling command‑line tools or multiple files.

Результат

Miso enables rapid, repeatable experimentation with multiple local music AI models, providing a unified workflow for generation, editing, stem separation, transcription, and remixing, all within a local Docker environment.

Рабочий процесс

• 1. Create a project and import a source audio file or generate a new take via a prompt.
• 2. Use the guided builder to set style, BPM, key, and other parameters for the chosen model.
• 3. Generate or repaint sections of a track using ACE‑Step or other models.
• 4. Split the track into stems (vocals, drums, bass, etc.).
• 5. Transcribe a stem to MIDI and convert to an ABC score with YuE2.
• 6. Edit the score or apply voice conversion, then remix and export the final take.
• 7. Save the project to preserve the lineage of prompts, takes, and edits.

Доказательства

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

Источник

The whole thing runs as two Docker containers.

Источник