MockingBird

MockingBird is a free, open-source voice-cloning toolbox that can copy a voice from just five seconds of audio and then speak arbitrary text in it, in real time. The project has earned more than 36,000 stars on GitHub and continues to be updated, with its latest changes in 2026. It remains one of the most popular self-hosted ways to experiment with few-shot text-to-speech.

Overview

MockingBird follows a classic three-stage pipeline: a speaker encoder extracts the identity of a voice from a short clip, a synthesizer converts text into a mel spectrogram in that voice, and a vocoder turns the spectrogram into listenable audio. A browser-based toolbox makes the workflow approachable — record or upload a sample, type your text and generate — while the underlying PyTorch code is fully available for developers to train, fine-tune and embed. The codebase has also become a common starting point for community forks and for research into few-shot speech synthesis. Trained voices work across the supported languages, and generation can keep up with real-time use cases on a decent GPU.

Key Features

• Clone a voice from as little as five seconds of reference audio
• Real-time speech generation in the cloned voice
• Web toolbox for recording, managing voices and synthesizing
• Modular encoder, synthesizer and vocoder stages you can retrain
• Pretrained models are provided so you can start cloning immediately
• Runs locally — voices and generated audio stay on your machine

💳 Pricing

MockingBird is free and open source, published on GitHub for anyone to download and run. There is no hosted service or paid tier; you supply a GPU-equipped machine and, optionally, your own training data.

Who Is It For

MockingBird is for developers, researchers and creators experimenting with voice cloning — prototyping dubbing, accessibility voices or character speech locally, before committing to a commercial voice platform.