app.py

# -*- coding: utf-8 -*-
import gradio as gr
import requests
import json
from pydub import AudioSegment
import os

# API Keys and Endpoints
TTS_URL = "https://talkbot.ir/TTS-tkun"
TEXT_GEN_URL = "https://talkbot.ir/api/v1/chat/completions"
TEXT_GEN_API_KEY = "sk-4fb613f56acfccf731e801b904cd89f5"
TEXT_GEN_MODEL = "deepseek-v3-0324"

def generate_podcast(topic):
    """
    Generates a podcast based on a given topic.
    """
    try:
        # 1. Generate podcast script using the text generation API
        print("Generating podcast script...")
        headers = {
            "Authorization": f"Bearer {TEXT_GEN_API_KEY}",
            "Content-Type": "application/json"
        }
        payload = {
            "model": TEXT_GEN_MODEL,
            "messages": [
                {"role": "system", "content": "به عنوان یک تولیدکننده پادکست، یک متن جذاب برای پادکست با دو گوینده (صدای اول و صدای دوم) در مورد موضوع زیر ایجاد کنید. متن باید شامل گفتگو بین دو گوینده باشد."},
                {"role": "user", "content": f"موضوع: {topic}"}
            ],
             "stream": False
        }

        response = requests.post(TEXT_GEN_URL, headers=headers, json=payload)
        response.raise_for_status()  # Raise an exception for bad status codes
        script_data = response.json()
        podcast_script = script_data['choices'][0]['message']['content']
        print("Podcast script generated.")

        # 2. Split the script into sections for each voice (This is a simplified approach)
        # You might need a more sophisticated parsing method for real-world scenarios
        lines = podcast_script.split('\n')
        voice1_lines = []
        voice2_lines = []
        current_voice = 1 # Assuming the first line is for voice 1

        for line in lines:
            if line.strip().startswith("صدای اول:"):
                voice1_lines.append(line.replace("صدای اول:", "", 1).strip())
                current_voice = 1
            elif line.strip().startswith("صدای دوم:"):
                voice2_lines.append(line.replace("صدای دوم:", "", 1).strip())
                current_voice = 2
            elif line.strip(): # Lines withoutexplicit voice indicators
                 if current_voice == 1:
                     voice1_lines.append(line.strip())
                 else:
                     voice2_lines.append(line.strip())


        voice1_text = " ".join(voice1_lines)
        voice2_text = " ".join(voice2_lines)

        if not voice1_text.strip() and not voice2_text.strip():
            return None, "خطا در تجزیه متن پادکست. لطفاً متن generado شده را بررسی کنید.", None

        # 3. Generate audio for each voice
        print("Generating audio for voice 1...")
        voice1_audio_url = requests.get(TTS_URL, params={"text": voice1_text}).text.strip()
        print(f"Voice 1 audio URL: {voice1_audio_url}")

        print("Generating audio for voice 2...")
        voice2_audio_url = requests.get(TTS_URL, params={"text": voice2_text}).text.strip()
        print(f"Voice 2 audio URL: {voice2_audio_url}")

        if not voice1_audio_url.startswith("http") or not voice2_audio_url.startswith("http"):
             return None, "خطا در تولید صدای TTS. لطفاً متن را بررسی کنید.", None

        # 4. Download audio files
        print("Downloading audio files...")
        voice1_audio_response = requests.get(voice1_audio_url)
        voice1_audio_response.raise_for_status()
        voice1_audio_path = "voice1.wav"
        with open(voice1_audio_path, "wb") as f:
            f.write(voice1_audio_response.content)

        voice2_audio_response = requests.get(voice2_audio_url)
        voice2_audio_response.raise_for_status()
        voice2_audio_path = "voice2.wav"
        with open(voice2_audio_path, "wb") as f:
             f.write(voice2_audio_response.content)

        print("Audio files downloaded.")

        # 5. Combine audio files (This is a simple concatenation for demonstration)
        # For a proper podcast with alternating speakers, you would need more complex pydub manipulation
        print("Combining audio files...")
        audio1 = AudioSegment.from_wav(voice1_audio_path)
        audio2 = AudioSegment.from_wav(voice2_audio_path)

        # A simple way to interleave: This assumes segments are short and alternate
        # For longer narrations, you'd need to split the audio files further
        combined_audio = AudioSegment.empty()
        max_len = max(len(audio1), len(audio2))
        segment_length = 5000 # Example segment length in milliseconds

        for i in range(0, max_len, segment_length):
            segment1 = audio1[i : i + segment_length]
            segment2 = audio2[i : i + segment_length]
            if segment1:
                combined_audio += segment1
            if segment2:
                combined_audio += segment2

        output_mp3_path = "podcast.mp3"
        combined_audio.export(output_mp3_path, format="mp3")
        print(f"Podcast saved as {output_mp3_path}")

        # 6. Clean up temporary files
        os.remove(voice1_audio_path)
        os.remove(voice2_audio_path)

        return podcast_script, None, output_mp3_path

    except requests.exceptions.RequestException as e:
        return None, f"خطا در اتصال به API ها: {e}", None
    except Exception as e:
        return None, f"خطای نامشخص: {e}", None

# Gradio Interface
def gradio_interface(topic):
    script, error, audio_path = generate_podcast(topic)
    if error:
        return error, None
    else:
        return script, audio_path

with gr.Blocks() as demo:
    gr.Markdown("## ساخت پادکست با هوش مصنوعی")
    topic_input = gr.Textbox(label="موضوع پادکست خود را وارد کنید")
    generate_button = gr.Button("شروع تولید پادکست")
    script_output = gr.Textbox(label="متن پادکست", interactive=False)
    audio_output = gr.Audio(label="پادکست تولید شده", type="filepath")

    generate_button.click(
        fn=gradio_interface,
        inputs=topic_input,
        outputs=[script_output, audio_output]
    )

if __name__ == "__main__":
    demo.launch()