from transformers import pipeline
import gradio as gr
import torch

pipe = pipeline(
    task="automatic-speech-recognition",
    model="iamgarvit/whisper-small-hi-asr",
    device=0 if torch.cuda.is_available() else -1,
)

def transcribe(audio_path):
    if audio_path is None:
        return ""
    result = pipe(audio_path)
    return result["text"]

demo = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(type="filepath", label="Speak or upload audio"),
    outputs=gr.Textbox(label="Transcription"),
    title="Whisper Small Hindi ASR",
    description=(
        "Hindi ASR (Automatic Speech Recognition) using a fine-tuned Whisper-small model. "
        "To use the microphone, click **Record** and allow microphone access when prompted "
        "by your browser. If microphone access is unavailable, you can upload an audio file instead."
    ),
)

demo.launch()