app.py

import os
import torch
from transformers import pipeline
import librosa
import gradio as gr
import warnings

# Suppress warnings for cleaner output
warnings.filterwarnings("ignore")

# ======================
# Model Loading
# ======================
def load_model():
    try:
        print("⏳ Loading Telugu ASR model...")
        # Using a model specifically fine-tuned for Telugu
        asr_pipeline = pipeline(
            "automatic-speech-recognition",
            model="anuragshas/wav2vec2-large-xlsr-53-telugu",
            device="cuda:0" if torch.cuda.is_available() else "cpu"
        )
        print("✅ Model loaded successfully")
        return asr_pipeline
    except Exception as e:
        print(f"❌ Model loading failed: {str(e)}")
        raise

# Load the ASR model
asr_pipeline = load_model()

# ======================
# Core Functionality
# ======================
def transcribe_telugu(audio_path):
    try:
        # Load audio file
        audio, sr = librosa.load(audio_path, sr=16000)
        
        # Perform transcription
        result = asr_pipeline(audio)
        telugu_transcription = result['text']
        
        return telugu_transcription

    except Exception as e:
        print(f"❌ Processing error: {str(e)}")
        return "Transcription failed (try clearer audio)"

# ======================
# Gradio Interface
# ======================
def create_interface():
    with gr.Blocks() as interface:
        gr.Markdown("""
        # 🎤 Telugu Speech Recognition
        Telugu Audio Transcription System:
        1. Speak in Telugu or upload an audio file
        2. Get accurate Telugu transcription

        Tips for best results:
        - Speak clearly in a quiet environment
        - Keep recordings between 5-30 seconds
        """)
        
        with gr.Row():
            audio_input = gr.Audio(
                type="filepath", 
                label="Record/Upload Telugu Audio",
                sources=["microphone", "upload"]
            )
        
        with gr.Row():
            output_text = gr.Textbox(label="Telugu Transcription")
        
        submit_btn = gr.Button("Transcribe")
        submit_btn.click(
            fn=transcribe_telugu,
            inputs=audio_input,
            outputs=output_text
        )
        
        gr.Examples(
            examples=[
                ["sample_telugu1.wav"],
                ["sample_telugu2.wav"]
            ],
            inputs=audio_input
        )
    
    return interface

# ======================
# Main Execution
# ======================
if __name__ == "__main__":
    print("🚀 Starting Telugu transcription service...")
    interface = create_interface()
    interface.launch(
        server_name="0.0.0.0",
        share=False,
        debug=True
    )