PowerBeef/Vocello

Vocello: a local, private voice studio for Apple Silicon. Write a script, pick or describe a voice, and generate speech on-device, faster than realtime on an 8 GB M2 Mac mini. Native Swift + MLX, no Python. Mac app out now, iPhone beta on TestFlight. (Formerly QwenVoice.)

What it solves

Vocello is a local-first, high-performance text-to-speech (TTS) studio for macOS and iOS. It eliminates the need for cloud-based credits, accounts, and external servers by running generative AI voice models directly on Apple Silicon hardware, ensuring that all scripts and recordings remain private on the user's device.

How it works

The application uses a native Swift runtime built on MLX (Apple's machine learning framework) to execute Qwen3-TTS models. Unlike many local TTS tools that rely on Python wrappers, Vocello uses a specialized first-party runtime (VocelloQwen3Core) to generate audio. It supports three primary workflows: using built-in voices, designing new voices via plain-language descriptions, and cloning existing voices from audio references.

Who it’s for

It is designed for users on Apple Silicon Macs (macOS 26+) and iPhone 15 Pro or newer (iOS 26+) who need high-quality, private voice generation for scripts, long-form projects, and custom voice personas.

Highlights

  • Local-First Privacy: All generation, history, and voice references are stored locally; no data leaves the device.
  • Voice Design & Cloning: Create voices from text descriptions or clone them from audio files/recordings.
  • High Performance: Optimized for Apple Silicon, achieving faster-than-realtime generation even on 8GB RAM devices.
  • Long-form Support: Handles scripts over 900 characters by streaming segments to maintain a flat memory footprint.
  • Multilingual: Supports ten languages including English, Chinese, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian.
  • Deterministic Output: Supports seed pinning for reproducible audio takes.

Related

  • Project
    soniqo/speech-swift

    A library of on-device AI speech models for Apple Silicon, providing local speech recognition, synthesis, and understanding for Mac and iOS apps.

  • Project
    Blaizzy/mlx-audio-swift

    A modular Swift SDK for audio processing on Apple Silicon, providing on-device support for text-to-speech, speech-to-text, and speaker diarization using the MLX framework.

  • Project
    Blaizzy/mlx-audio

    MLX‑Audio is a Python library (with CLI, web UI, and Swift package) that runs a large collection of modern speech and music models on Apple Silicon via the MLX framework. It supports TTS, STT, speech‑to‑speech, voice cloning, VAD/diarisation, and music generation, offering quantised 3‑ to 8‑bit variants for low‑memory inference. Install with `pip install mlx-audio`, generate audio via `mlx_audio.tts.generate …` or the Python `load_model(...).generate` API, and optionally serve an OpenAI‑compatible REST API.

  • Project
    james-see/ltx-video-mac

    A native macOS application for generating AI videos with synchronized audio from text or images using the LTX-2 model and MLX framework on Apple Silicon.

  • Project
    Blaizzy/nativ

    A native macOS app for running MLX models locally on Apple silicon, featuring a chat interface, image generation, and an OpenAI-compatible API server.