tests/tools/test_whisper.py

import importlib.resources
import json

import torch

from pathlib import Path
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline

from question_retriever import get_question
from tools.data_helpers import get_file_path

__resources_path = Path(str(importlib.resources.files("data")))


def test_whisper() -> None:

    task_id = "1f975693-876d-457b-a649-393859e79bf3"
    question = json.loads(get_question(task_id=task_id))

    audio_file = get_file_path(file_name=question["file_name"])

    # cuda_available = torch.cuda.is_available()
    cuda_available = False
    device = "cuda:0" if cuda_available else "cpu"
    torch_dtype = torch.float16 if cuda_available else torch.float32

    model_id = "openai/whisper-large-v3-turbo"

    model = AutoModelForSpeechSeq2Seq.from_pretrained(
        model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
    )
    model.to(device)

    processor = AutoProcessor.from_pretrained(model_id)

    pipe = pipeline(
        "automatic-speech-recognition",
        model=model,
        tokenizer=processor.tokenizer,
        feature_extractor=processor.feature_extractor,
        torch_dtype=torch_dtype,
        device=device,
    )

    sample = audio_file

    generate_kwargs = {
        "return_timestamps": True,
    }

    result = pipe(sample, generate_kwargs=generate_kwargs)

    print(result["text"])