import mlx_whisper

result = mlx_whisper.transcribe(
    "audio.wav",
    path_or_hf_repo="Leonidng/kb-whisper-small-mlx",
    language="sv",
)
print(result["text"])

Model details

Base model: KBLab/kb-whisper-small (Swedish fine-tuned Whisper small)
Format: MLX (float16 safetensors)
Size: ~459 MB
Conversion: Converted using mlx-examples/whisper/convert.py

KBLab's Swedish fine-tune of Whisper small outperforms OpenAI's whisper-large-v3 on Swedish benchmarks while being significantly smaller and faster.