| from pprint import pprint | |
| from transformers import pipeline | |
| from datasets import load_dataset | |
| # config | |
| model_id = "kotoba-tech/kotoba-whisper-v1.0" | |
| generate_kwargs = {"language": "japanese", "task": "transcribe"} | |
| # load model | |
| pipe = pipeline( | |
| "automatic-speech-recognition", | |
| model=model_id, | |
| chunk_length_s=15, | |
| batch_size=64 | |
| ) | |
| # load sample audio (concatenate instances to create a long audio) | |
| dataset = load_dataset("kotoba-tech/kotoba-whisper-eval", split="train") | |
| x = dataset['audio'][0] | |
| elapsed = {} | |
| for x in dataset['audio']: | |
| start = time() | |
| transcription = pipe(x.copy(), generate_kwargs=generate_kwargs) | |
| elapsed[x['path']] = time() - start | |
| pprint(elapsed) |