---
language:
  - ja
pipeline_tag: automatic-speech-recognition
library_name: transformers.js
base_model: UsefulSensors/moonshine-tiny-ja
tags:
  - moonshine
  - onnx
  - transformers.js
  - automatic-speech-recognition
  - japanese
license: other
license_name: moonshine-ai-community-license
license_link: https://github.com/usefulsensors/moonshine/blob/main/LICENSE
---

# Moonshine Tiny JA (ONNX / Transformers.js)

[UsefulSensors/moonshine-tiny-ja](https://huggingface.co/UsefulSensors/moonshine-tiny-ja) の ONNX 版です。Transformers.js / onnxruntime-web で使用できます。

## Demo

[Moonshine Tiny JA Demo](https://huggingface.co/spaces/wmoto-ai/moonshine-tiny-ja-demo)

## Usage

```javascript
import {
  MoonshineForConditionalGeneration,
  AutoProcessor,
  AutoTokenizer,
} from '@huggingface/transformers';

const MODEL_ID = 'wmoto-ai/moonshine-tiny-ja-ONNX';

const [model, processor, tokenizer] = await Promise.all([
  MoonshineForConditionalGeneration.from_pretrained(MODEL_ID, { dtype: 'fp32' }),
  AutoProcessor.from_pretrained(MODEL_ID),
  AutoTokenizer.from_pretrained(MODEL_ID),
]);

// audioData: Float32Array (16kHz, mono)
const inputs = await processor(audioData);
const outputs = await model.generate({ ...inputs, max_new_tokens: 64 });
const text = tokenizer.decode(outputs[0], { skip_special_tokens: true });
```

## Changes from Original

- `attention_mask` 入力を削除（Transformers.js互換のため内部生成に変更）
- Transformers.js が期待するディレクトリ構造に配置

重み自体は元モデルと同一です。

## Intended use / Limitations

- 主用途: 日本語音声のリアルタイム文字起こし
- 非推奨: 医療・法務など高リスク領域での利用
- 詳細は[元モデル](https://huggingface.co/UsefulSensors/moonshine-tiny-ja)を参照

## License

[Moonshine AI Community License](https://github.com/usefulsensors/moonshine/blob/main/LICENSE) に準拠します。詳細は元モデルおよびライセンス全文を参照してください。

## Citation

```bibtex
@misc{king2025flavorsmoonshinetinyspecialized,
      title={Flavors of Moonshine: Tiny Specialized ASR Models for Edge Devices},
      author={Evan King and Adam Sabra and Manjunath Kudlur and James Wang and Pete Warden},
      year={2025},
      eprint={2509.02523},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2509.02523}
}
```