app.py

import modal

app = modal.App("whisper-app")

# Define container with required packages
image = (
    modal.Image.debian_slim()
    .pip_install("openai-whisper", "requests")
    .apt_install("ffmpeg")
)

@app.cls(image=image)
class WhisperModel:
    def __init__(self):
        self.model = None

    @modal.enter()  # ✅ This is critical: use @modal.enter() not just __enter__
    def init(self):
        import whisper
        self.model = whisper.load_model("base")

    @modal.method()
    def transcribe(self, audio_url):
        import requests

        # Download audio file
        response = requests.get(audio_url)
        with open("audio.wav", "wb") as f:
            f.write(response.content)

        # Run transcription
        result = self.model.transcribe("audio.wav")
        return result["text"]

@app.local_entrypoint()
def main():
    url = "https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/mlk.flac"

    model = WhisperModel()
    result = model.transcribe.remote(url)
    print(result)