Hysen LabsMockingBird
🚀Clone a voice in 5 seconds to generate arbitrary speech in real-time
MockingBird clones a voice in five seconds, with caveats attached
The Chinese-focused voice cloning toolbox still works, but its author has moved on to building a hosted product, and the README is upfront about that.
The claim on the box
Clone a voice in five seconds and generate arbitrary speech in real time: that is the repository's own summary, and the toolbox inside is what backs it. The code is Python, the repo holds near 37,000 stars, and it remains available as an open source reference even as active development has wound down.
Chinese first
Chinese is the language with real testing behind it. The README lists work with the aidatatang 200zh, magicdata, aishell3, and data aishell datasets, a set of Mandarin corpora. That focus positions the tool for Chinese speech synthesis while leaving other languages more of an experiment than a promise.
What you need to run it
Python 3.7 or higher, plus PyTorch, tested with version 1.9.0 and GPU hardware like a Tesla T4 or a GTX 2060. An easy path to decent output is documented: reuse the pretrained encoder and vocoder, and train only a new synthesizer. For environment checks, the general setup section points back to the original repository.
The state of the project
The author states plainly that the repository is no longer actively updated. The technology is still being pushed forward in open source elsewhere, and an optimized, cloud hosted version is being built at a separate site, with a mention of hiring. Read that as you would any paused project: usable, popular, and unlikely to receive fixes.
Editorial conclusion
If your use case is Chinese synthesis and the pinned versions work for you, there is still value here. Just go in knowing the clock stopped.
Community notes