Your Mac, a private voice cloning studio

What it does

MimikaStudio is a local-first macOS app for Apple Silicon that clones voices, generates text-to-speech, reads documents aloud, and turns full books into audiobooks. It runs entirely on-device via MLX native acceleration and exposes both a Flutter UI and an API for automation. The app also acts as an agentic voice cloning server with job queue orchestration for batch TTS, cloning, and audiobook pipelines.

The interesting bit

The project bundles a surprising variety of open-source audio models under one roof—Kokoro for fast English TTS, Qwen3-TTS and Chatterbox for multilingual voice cloning from 3-second samples, and Supertonic for ONNX-based synthesis—then wraps them in a queueable, MCP-aware server so other local agents can submit voice jobs without touching the cloud.

Key highlights

• Clones voices from as little as 3 seconds of reference audio using Qwen3-TTS or Chatterbox, with 8 bundled reference voices ready to go.
• Reads PDF, DOCX, EPUB, Markdown, and TXT aloud with sentence-level highlighting and synchronized progression.
• Generates queueable audiobooks with chapter-level batching and reusable voice presets.
• Runs fully offline on Apple Silicon with MLX Metal acceleration; models are downloaded on first run.
• Exposes an agentic MCP dashboard and API for local automation workflows.

Caveats

• macOS binaries only; the codebase is cross-platform but Windows builds are not yet provided.
• Licensed under BSL-1.1 for source and a separate binary distribution license, so check the terms before redistributing.

Verdict

Worth a look if you want a private, offline audio pipeline on a Mac. Windows users and anyone looking for a simple open-source license should wait or look elsewhere.