Parakeet-Inc/furigana_whisper_small_jsut

from transformers import pipeline
from pathlib import Path

pipe = pipeline(
    "automatic-speech-recognition",
    model="Parakeet-Inc/furigana_whisper_small_jsut",
)

def transcribe_with_prompt(pipe, audio_path: str | Path, prompt: str) -> str:
    prompt_ids = pipe.tokenizer.get_prompt_ids(
        prompt, return_tensors="pt"
    ).to(pipe.device)
    generate_kwargs = {"prompt_ids": prompt_ids}
    result = pipe(str(audio_path), generate_kwargs=generate_kwargs)
    return result["text"]

# 実行例
audio_path = "path/to/your/audio.wav"
prompt = "明日は晴れ。"
transcription = transcribe_with_prompt(pipe, audio_path, prompt)
print(transcription)  # アスワハレ。