Audio8-TTS-Preview-0.6b
Generates multilingual speech and clones voices from audio samples using a compact open-source model.
The gist
Audio8-TTS-Preview-0.6b is a compact, 0.6 billion-parameter text-to-speech model from the Audio8 team. It solves the problem of generating high-quality, multilingual speech and performing zero-shot voice cloning with a smaller model footprint compared to state-of-the-art competitors. It currently supports 11 languages, with broader coverage planned for future releases.
What it does
- Generates speech in 11 languages, including Cantonese, Chinese, Dutch, English, and French.
- Clones a speaker's voice from a short reference audio clip (zero-shot).
- Produces high-fidelity 44.1 kHz audio output.
- Operates as a compact 0.6B parameter model, smaller than many leading alternatives.
- Decodes generated audio codes into standard waveform files like WAV.
How it works
Users interact with the model via Python, using the Transformers library and providing trust_remote_code=True. For voice cloning, you input a reference audio file and its exact transcript alongside the target text. The model outputs a synthesized audio waveform. Released under the Apache 2.0 license, it is designed to be self-hosted on a CUDA-capable GPU.
Best for
This model is best for developers and researchers who need to integrate high-quality multilingual TTS and voice cloning into applications where computational efficiency and a smaller model size are important.
Watch out for
This is a preview release with limited language coverage. Voice cloning performance can degrade if reference audio clips are long, noisy, or incorrectly transcribed. The model can be misused for impersonation, so consent and disclosure are critical.