---
license: apache-2.0
tags:
- moss-tts
- tts
- text-to-speech
- rust
- candle
- audio
---

# MOSS-TTS-Nano (Rust/Candle)

Rust/Candle implementation of [MOSS-TTS-Nano](https://github.com/OpenMOSS/MOSS-TTS-Nano), 
a 100M-parameter multilingual TTS model with voice cloning support.

## Model Files

| File | Description |
|------|-------------|
| `config.json` | MOSS-TTS-Nano LM config (HuggingFace-compatible) |
| `tokenizer.model` | SentencePiece tokenizer |
| `moss_tts_nano_lm.safetensors` | Language model weights (F32, 285 MB) |
| `moss_audio_tokenizer.safetensors` | Audio tokenizer weights (F32, merged weight-norm, 88 MB) |

## Quick Start

```rust
// Cargo.toml
// hf-hub = "0.3"

use hf_hub::hf_hub_download;

// Download all model files at once
let lm_repo = "ramishi/moss-tts-nano-candle";
let lm_weights = hf_hub_download(lm_repo, "moss_tts_nano_lm.safetensors")?;
let config    = hf_hub_download(lm_repo, "config.json")?;
let tokenizer = hf_hub_download(lm_repo, "tokenizer.model")?;
let codec_weights = hf_hub_download(lm_repo, "moss_audio_tokenizer.safetensors")?;
let codec_config  = hf_hub_download(lm_repo, "config.json")?; // shared config
```

## Credits

Based on [OpenMOSS-Team/MOSS-TTS-Nano-100M](https://huggingface.co/OpenMOSS-Team/MOSS-TTS-Nano-100M).