MARS5 TTS

MARS5 TTS's 概要

MARS5 is a novel English speech model (TTS) developed by CAMB.AI. It follows a two-stage AR-NAR pipeline with a unique NAR component. With just 5 seconds of audio and a snippet of text, MARS5 can generate speech for diverse scenarios like sports commentary and anime. The model can be steered with punctuation and capitalization to guide the prosody of the generated output. Speaker identity can be specified using an audio reference file. MARS5 supports both shallow and deep cloning, with the latter requiring the prompt transcript. The model can be easily loaded using `torch.hub` and inference can be performed by providing the reference audio and transcript. The default settings provide good results, but the inference settings can be tuned for specific use cases. The checkpoints for MARS5, along with the necessary hardware requirements, are provided on the GitHub repo. Contributions to improve the model are welcome.

MARS5 TTS's 特徴

• Two-stage AR-NAR pipeline

• Guided prosody using punctuation and capitalization

• Speaker identity specification

• Shallow and deep cloning

• Easy model loading using `torch.hub`

• Inference using reference audio and transcript

• Open-source with alternative licensing options

MARS5 TTS's Q&A

MARS5 TTS's 価格

MARS5 is open-source and available under GNU AGPL 3.0 license. For alternative licensing options, please contact [email protected]