maya-research/maya2-native
AI model · https://huggingface.co/maya-research/maya2-native
First seen by Oriel on .
AI summary
Maya 2 Native is an Apache-2.0 streaming text-to-speech model generating 24 kHz speech across 12 languages, with 20 voices, cross-lingual voice cloning, and self-hosted inference interfaces.
- The 2.50B-parameter model provides 20 built-in voices, zero-shot cloning from 3–10 seconds of reference audio, and supports Assamese, Bengali, English, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, and Telugu.
- Open weights and inference code include a Python API, OpenAI-compatible HTTP speech endpoint, WebSocket streaming, continuous batching, and CUDA graphs for concurrent generation.
- The documented single-seat loading peak is approximately 9.2 GB plus CUDA context; Linux, Python 3.10–3.12, and an NVIDIA GPU supporting CUDA 12.8 are required.
- On 6,000 held-out texts, overall character error rates were 3.49%, 3.40%, and 2.86% under scorer v1 across three ASR judges, while voice identity reached 98.6%.
- Voice cloning can produce speech in a language different from the reference, but similarity and accent vary; the release includes no dedicated cloning benchmark and the server has no built-in authentication.
#text-to-speech #voice cloning #multilingual speech #cuda
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Likes
Total likes reported by the source.
- Hugging Face Models: 151, latest observation ; +42 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Models: 3,810, latest observation ; +3,466 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | maya-research/maya2-native | https://huggingface.co/maya-research/maya2-native Read discussion |