---
license: mit
language:
- en
library_name: onnx-asr
tags:
- onnx
- onnx-asr
- automatic-speech-recognition
- moonshine
base_model: UsefulSensors/moonshine-tiny
---

# Moonshine Tiny (ONNX)

Moonshine Tiny is a compact English speech-to-text model from
[Useful Sensors](https://github.com/usefulsensors/moonshine). It is an
encoder-decoder model that reads the raw 16 kHz waveform, so it needs no
log-mel preprocessor.

This repository is a mirror of the official ONNX export
[onnx-community/moonshine-tiny-ONNX](https://huggingface.co/onnx-community/moonshine-tiny-ONNX),
kept so that the OpenVoiceOS ONNX ASR collection is self-contained. The
weights are unchanged.

## Usage

```py
import onnx_asr

model = onnx_asr.load_model("moonshine-tiny")
print(model.recognize("test.wav"))
```

Quantized variants load with the `quantization` argument:

```py
model = onnx_asr.load_model("moonshine-tiny", quantization="quantized")
```

## Credits

Model by Useful Sensors Inc., released under the MIT license. Paper:
[Moonshine: Speech Recognition for Live Transcription and Voice Commands](https://arxiv.org/abs/2410.15608).
ONNX export by [onnx-community](https://huggingface.co/onnx-community).

## License

MIT