README.md

---

license: apache-2.0
language:
- ja
library_name: mlx
tags:
- whisper
---


# kotoba-whisper-v1.0-mlx

This repository contains a converted `mlx-whisper` model of [kotoba-whisper-v1.0](https://huggingface.co/kotoba-tech/kotoba-whisper-v1.0) which is suitable for running with Apple Silicon.
As `kotoba-whisper-v1.0` is derived from `distil-large-v3`, this model is significantly faster than [mlx-community/whisper-large-v3-mlx](https://huggingface.co/mlx-community/whisper-large-v3-mlx) without losing much accuracy for Japanese transcription.

## Usage

```sh
pip install mlx-whisper
```

```py
mlx_whisper.transcribe(speech_file, path_or_hf_repo="kaiinui/kotoba-whisper-v1.0-mlx")
```

## Related Links
* [kotoba-whisper-v1.0](https://huggingface.co/kotoba-tech/kotoba-whisper-v1.0) (The original model)
* [mlx-whisper](https://github.com/ml-explore/mlx-examples/tree/main/whisper)