app.py

import gradio as gr
import torch
import numpy as np
from transformers import WhisperProcessor, WhisperForConditionalGeneration
from googletrans import Translator
from pydub import AudioSegment
import io

# Load the Whisper model and processor once
processor = WhisperProcessor.from_pretrained("openai/whisper-large-v2")
model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large-v2")

def transcribe_and_translate(audio, target_language=None):
    try:
        # Convert the Gradio audio input to an AudioSegment
        audio_segment = AudioSegment.from_file(io.BytesIO(audio), format="wav")
        audio_segment = audio_segment.set_channels(1).set_frame_rate(16000)
        
        # Prepare audio for Whisper
        buffer = io.BytesIO()
        audio_segment.export(buffer, format="wav")
        waveform = torch.FloatTensor(np.frombuffer(buffer.getvalue(), dtype=np.int16) / 32768.0).unsqueeze(0)

        # Transcription
        inputs = processor(waveform, return_tensors="pt", sampling_rate=16000)
        with torch.no_grad():
            logits = model.generate(inputs["input_features"])
        transcription = processor.batch_decode(logits, skip_special_tokens=True)[0]

        if target_language:
            translator = Translator()
            translation = translator.translate(transcription, dest=target_language)
            return transcription, translation.text
        else:
            return transcription, None

    except Exception as e:
        return f"An error occurred: {e}", None

# Gradio interface
def process_audio(audio, target_language):
    transcription, translation = transcribe_and_translate(audio, target_language)
    if translation:
        return transcription, translation
    else:
        return transcription, "No translation requested."

# Define the Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("## Audio Transcription and Translation")
    
    audio_input = gr.Audio(source="microphone", type="file", label="Record or Upload Audio")
    target_language = gr.Textbox(label="Target Language (e.g., 'es' for Spanish)", placeholder="Leave blank for no translation")

    transcription_output = gr.Textbox(label="Transcription")
    translation_output = gr.Textbox(label="Translation")

    submit_button = gr.Button("Transcribe and Translate")

    submit_button.click(process_audio, inputs=[audio_input, target_language], outputs=[transcription_output, translation_output])

# Launch the Gradio interface
demo.launch()