app.py

import gradio as gr
import torch
import soundfile as sf
from transformers import pipeline

device = "cuda:0" if torch.cuda.is_available() else "cpu"
pipe = pipeline(
    "automatic-speech-recognition",
    model="vrclc/W2V2-BERT-withLM-Malayalam-Studio",
    chunk_length_s=10,
    device=device,
)

def transcribe(audio):
    """Transcribes Malayalam speech from an audio file."""
    try:
        if audio is None:
            return "Please record or upload an audio file."

        print(f"[DEBUG] Received audio: {audio}")

        # Handle filepath case from Gradio
        audio_path = audio if isinstance(audio, str) else audio.get("name", None)
        if audio_path is None:
            return "Could not read audio file."

        print(f"[DEBUG] Reading audio file: {audio_path}")
        audio_data, sample_rate = sf.read(audio_path)

        print(f"[DEBUG] Audio sample rate: {sample_rate}, shape: {audio_data.shape}")

        transcription = pipe(
            {"array": audio_data, "sampling_rate": sample_rate},
            chunk_length_s=10,
            batch_size=8,
        )["text"]

        print(f"[DEBUG] Transcription: {transcription}")
        return transcription

    except Exception as e:
        import traceback
        print("[ERROR] Exception during transcription:")
        traceback.print_exc()
        return f"Error: {str(e)}"

iface = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(sources=["microphone", "upload"], type="filepath"),
    outputs="text",
    title="Malayalam Speech Recognition using w2v2-BERT with LM",
    description="Record or upload Malayalam speech and submit to get the transcribed text.",
    examples=[["sample.wav"]],
)
iface.launch()