MLX-Audio

MLX-Audio is an open-source speech library built on Apple’s MLX framework that brings text-to-speech (TTS), speech-to-text (STT) and speech-to-speech (STS) to Apple Silicon Macs. Instead of sending audio to a cloud API, it runs modern speech models locally on the GPU and Neural Engine through MLX, so generation is fast, private and free of per-request fees. The project has earned about 8,100 stars on GitHub and is actively maintained, with updates landing in 2026.

Overview

The library gives developers one consistent Python API and command-line interface for a growing collection of state-of-the-art open speech models. You can synthesize natural-sounding voices, transcribe recordings, or chain the two together for speech-to-speech conversion, and models can be converted into MLX format and quantized to run efficiently within the unified memory of M-series chips. Because everything executes on-device, MLX-Audio fits workflows where latency, privacy or offline operation matter: voice interfaces, accessibility tooling, podcast and video narration, and realtime assistants on a Mac. It builds directly on the MLX ecosystem from Apple’s machine learning research team, so models load quickly and share memory with the rest of your MLX pipeline rather than needing a separate runtime.

Key Features

• Text-to-speech, speech-to-text and speech-to-speech in a single library
• Runs entirely on Apple Silicon via the MLX framework — no cloud API needed
• Supports a growing roster of popular open speech models for generation and transcription
• Convert and quantize models to MLX format for lower memory use and faster inference
• Simple Python API plus a command-line tool for quick experiments
• Streaming-friendly generation for low-latency voice applications

💳 Pricing

MLX-Audio is completely free and open source under the MIT license. You only need your own Mac with Apple Silicon; the speech models it runs are open weights.

Who Is It For

MLX-Audio is for Mac developers and researchers who want high-quality speech synthesis and transcription running locally on Apple Silicon, without cloud costs or sending audio to third-party services.