这是
kotoba-tech/kotoba-whisper-v2.1 的
MLX 格式转换版本,专为 Apple Silicon (M1/M2/M3/M4) 上的 MLX 推理优化。
1mlx_whisper --model enet45/kotoba-whisper-v2.1-mlx-fp16 \
2 --language ja \
3 --task transcribe \
4 --output-format srt \
5 /path/to/your/audio.wav
1from mlx_whisper import transcribe
2
3result = transcribe(
4 "/path/to/audio.wav",
5 path_or_hf_repo="enet45/kotoba-whisper-v2.1-mlx-fp16",
6 language="ja",
7 task="transcribe",
8)
9
10print(result["text"])
11
12# 带时间戳的分段
13for seg in result["segments"]:
14 print(f"[{seg['start']:.1f}s - {seg['end']:.1f}s] {seg['text']}")
1@misc{kotoba-whisper-v2.1,
2 title={Kotoba-Whisper},
3 author={Kotoba Technologies},
4 year={2025},
5 publisher={Hugging Face},
6 url={https://huggingface.co/kotoba-tech/kotoba-whisper-v2.1}
7}
