app.py

# app.py

import gradio as gr
import torch
import torchaudio
import google.generativeai as genai
from transformers import AutoProcessor, AutoModel
import numpy as np
import os
import json

# Initialize Gemini AI
genai.configure(api_key='YOUR_GEMINI_API_KEY')
model = genai.GenerativeModel('gemini-pro')

# Initialize F5-TTS model
processor = AutoProcessor.from_pretrained("SWivid/F5-TTS")
f5_model = AutoModel.from_pretrained("SWivid/F5-TTS")

def clone_voice(audio_file):
    waveform, sample_rate = torchaudio.load(audio_file)
    # Implement voice cloning logic here
    # For demonstration, we'll just return a placeholder
    return "Voice cloned successfully"

def generate_podcast_script(content, duration):
    prompt = f"""
    Create a podcast script for two people discussing the following content:
    {content}
    
    The podcast should last approximately {duration}. Include natural speech patterns,
    humor, and occasional off-topic chit-chat. Use speech fillers like "um", "ah",
    "yes", "I see", "Ok now". Vary the emotional tone (e.g., regular, happy, sad, surprised)
    and indicate these in [square brackets]. Format the script as follows:

    Host 1: [emotion] Dialog
    Host 2: [emotion] Dialog
    
    Ensure the conversation flows naturally and stays relevant to the topic.
    """
    response = model.generate_content(prompt)
    return response.text

def text_to_speech(text, speaker_id):
    inputs = processor(text=text, return_tensors="pt")
    speech = f5_model.generate_speech(inputs["input_ids"], speaker_id=speaker_id, vocoder_output=True)
    return speech.cpu().numpy()

def create_podcast(content, duration, voice1, voice2):
    script = generate_podcast_script(content, duration)
    lines = script.split('\n')
    audio_segments = []
    
    for line in lines:
        if line.startswith("Host 1:"):
            audio = text_to_speech(line[7:], speaker_id=0)  # Assuming speaker_id 0 for Host 1
            audio_segments.append(audio)
        elif line.startswith("Host 2:"):
            audio = text_to_speech(line[7:], speaker_id=1)  # Assuming speaker_id 1 for Host 2
            audio_segments.append(audio)
    
    # Concatenate audio segments
    podcast_audio = np.concatenate(audio_segments)
    return (22050, podcast_audio)  # Assuming 22050 Hz sample rate

def gradio_interface(content, duration, voice1, voice2):
    script = generate_podcast_script(content, duration)
    return script

def render_podcast(script, voice1, voice2):
    lines = script.split('\n')
    audio_segments = []
    
    for line in lines:
        if line.startswith("Host 1:"):
            audio = text_to_speech(line[7:], speaker_id=0)
            audio_segments.append(audio)
        elif line.startswith("Host 2:"):
            audio = text_to_speech(line[7:], speaker_id=1)
            audio_segments.append(audio)
    
    podcast_audio = np.concatenate(audio_segments)
    return (22050, podcast_audio)

# Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("# AI Podcast Generator")
    
    with gr.Row():
        content_input = gr.Textbox(label="Paste your content or upload a document")
        document_upload = gr.File(label="Upload Document")
    
    duration = gr.Radio(["1-5 min", "5-10 min", "10-15 min"], label="Estimated podcast duration")
    
    with gr.Row():
        voice1_upload = gr.Audio(label="Upload Voice 1", type="filepath")
        voice2_upload = gr.Audio(label="Upload Voice 2", type="filepath")
    
    generate_btn = gr.Button("Generate Script")
    script_output = gr.Textbox(label="Generated Script", lines=10)
    
    render_btn = gr.Button("Render Podcast")
    audio_output = gr.Audio(label="Generated Podcast")
    
    generate_btn.click(gradio_interface, inputs=[content_input, duration, voice1_upload, voice2_upload], outputs=script_output)
    render_btn.click(render_podcast, inputs=[script_output, voice1_upload, voice2_upload], outputs=audio_output)

demo.launch()