---
license: other
license_name: apache-2.0-and-mit
tags:
  - audio.cpp
  - gguf
  - text-to-speech
  - voice-design
---

# Maya1 GGUF

For use with [audio.cpp](https://github.com/0xShug0/audio.cpp).

This package combines the autoregressive weights from
[maya-research/maya1](https://huggingface.co/maya-research/maya1/tree/21c682a0afef8c13a89b2512733c8bf5f0c52eb7)
with the decoder and quantizer from
[hubertsiuzdak/snac_24khz](https://huggingface.co/hubertsiuzdak/snac_24khz/tree/d73ad176a12188fcf4f360ba3bf2c2fbbe8f58ec).

`maya1-orig.gguf` preserves the original mixed precision: BF16 Maya1 weights
and FP32 SNAC weights. It also embeds the tokenizer, model configuration, and
audio.cpp model spec, so no separate runtime files are required.

`maya1-q8_0.gguf` stores eligible weights as Q8_0. It is also self-contained
and can be selected directly in the same CLI, server, and WebUI workflows.

## Usage

```bash
audiocpp_cli --task tts --family maya1 \
  --model Maya1-GGUF/maya1-orig.gguf \
  --backend cuda \
  --text "Welcome back. <laugh> Today is going to be interesting." \
  --request-option 'instruct=Warm female voice in her 30s with an American accent and conversational pacing.' \
  --out maya1.wav --log
```

See the [Maya1 usage guide](https://github.com/0xShug0/audio.cpp/blob/main/docs/models/maya1.md)
for sampling and long-form options.

## License

The Maya1 weights are distributed under Apache-2.0; see `LICENSE-MAYA1`.
The bundled SNAC weights are distributed under MIT; see `LICENSE-SNAC`.
Conversion to GGUF does not replace either upstream license.