app.py

import gradio as gr
from transformers import pipeline
import torch

# Load the Whisper model pipeline for speech recognition with optimizations
model_name = "Vira21/Whisper-Small-Khmer"
whisper_pipeline = pipeline(
    "automatic-speech-recognition", 
    model=model_name,
    device=0 if torch.cuda.is_available() else -1  # Use GPU if available, otherwise use CPU
)

def transcribe_audio(audio):
    print(f"Received audio type: {type(audio)}")  # Debugging line
    try:
        result = whisper_pipeline(audio)["text"]
        return result
    except Exception as e:
        return f"An error occurred during transcription: {str(e)}"

# Gradio Interface with optimizations
interface = gr.Interface(
    fn=transcribe_audio,
    inputs=gr.Audio(type="filepath"),  
    outputs="text",
    title="Whisper Small Khmer-Eng Speech-to-Text",
    description="Upload an audio file or record your voice to get the transcription in Khmer-English.",
    examples=[["Example Audio/126.wav"], ["Example Audio/232.wav"], ["Example Audio/tomholland28282.wav"]],
    allow_flagging="never"  # Disables flagging to save resources
)

# Launch the app with queue enabled for better handling on free CPU
if __name__ == "__main__":
    interface.queue()  # Enable asynchronous queuing for better performance
    interface.launch()