Views
No views yet
| 指標 | Base Whisper | This LoRA | 改善 |
|---|---|---|---|
| Char WER | 8.61% | 5.59% | -35% 相対 |
| FIR (Filler Inclusion Rate) | 13.3% | 48.9% | 3.7倍 |
| filler recall | 6/45 | 22/45 | +16 |
1import torch
2from transformers import WhisperForConditionalGeneration, WhisperProcessor
3from peft import PeftModel
4
5processor = WhisperProcessor.from_pretrained(
6 "openai/whisper-large-v3", language="japanese", task="transcribe"
7)
8base = WhisperForConditionalGeneration.from_pretrained(
9 "openai/whisper-large-v3", dtype=torch.float16
10)
11model = PeftModel.from_pretrained(base, "Coidemo/whisper-large-v3-filler-lora")
12# 推論用にマージしたい場合:
13merged = model.merge_and_unload()Coidemo/whisper-large-v3-filler-mlx で公開されている。1import mlx_whisper
2
3result = mlx_whisper.transcribe(
4 "video.mp4",
5 path_or_hf_repo="Coidemo/whisper-large-v3-filler-mlx",
6 language="ja",
7)openai/whisper-large-v3core.mlx_whisper_refine.transcribe_refined で境界重複 dedup + hallucination retry を実装済みCoidemo/whisper-large-v3-filler-mlx1@article{radford2022whisper,
2 title={Robust Speech Recognition via Large-Scale Weak Supervision},
3 author={Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya},
4 journal={arXiv preprint arXiv:2212.04356},
5 year={2022}
6}