---
license: mit
pipeline_tag: text-to-speech
tags:
- irodori-tts
- text-to-speech
- japanese
- onnx
- migraphx
- rocm
- voice-design
---

# irodori_tts_cpp_artifacts

Exported ONNX and MIGraphX artifacts for running `Aratako/Irodori-TTS-500M-v2-VoiceDesign` with the native `irodori_tts_cpp` runtime.

This repository contains only the inference artifacts required by the C++ runtime:

- `manifest.json`
- local tokenizer JSON
- RF-DiT context ONNX / external data / MIGraphX cache
- RF-DiT step ONNX / external data / MIGraphX cache for 1, 2, 4, 8, 12, 16, 24, and 30 second buckets
- DACVAE decode ONNX / external data / MIGraphX cache for 1, 2, 4, 8, 12, 16, 24, and 30 second buckets

The runtime request path uses the `.mxr` files. ONNX and `.onnx.data` files are included as the cache generation source artifacts.

## Download

```bash
huggingface-cli download yoshou/irodori_tts_cpp_artifacts \
  --local-dir artifacts/irodori-500m-v2-voicedesign-test
```

## License

The exported Irodori-TTS VoiceDesign artifacts are distributed under the MIT License, following the upstream model license.

This repository also includes tokenizer files derived from `llm-jp/llm-jp-3-150m`, which is licensed under the Apache License 2.0. See `NOTICE` for attribution details.

Users must also follow the ethical restrictions described in the upstream Irodori-TTS VoiceDesign model card, including no impersonation and no misleading synthetic speech.

## Upstream

- https://huggingface.co/Aratako/Irodori-TTS-500M-v2-VoiceDesign
- https://github.com/Aratako/Irodori-TTS
- https://huggingface.co/llm-jp/llm-jp-3-150m