whisper_mac.py

import json
import time
import torch
from transformers import pipeline

pipe = pipeline(
    "automatic-speech-recognition",
    "openai/whisper-large-v3",
    torch_dtype=torch.float16,
    device="mps",
)

start = time.time()

output = pipe(
    "https://huggingface.co/datasets/reach-vb/random-audios/resolve/main/ted_60.wav",
    chunk_length_s=30,
    batch_size=24,
    return_timestamps=True,
)

end = time.time()
print("Total time!")
print(end - start)

with open("output.json", "w", encoding="utf8") as fp:
    json.dump(output, fp, ensure_ascii=False)