app.py

import gradio as gr
from pydub import AudioSegment
from groq import AsyncGroq
import json
import uuid
import io
import edge_tts
import asyncio
import aiofiles
import PyPDF2
import os
from typing import List, Dict, Tuple

class PodcastGenerator:
    def __init__(self, groq_api_key: str):
        self.groq_client = AsyncGroq(api_key=groq_api_key)

    async def generate_script(self, prompt: str) -> Dict:
        example = """
{
    "topic": "AGI",
    "podcast": [
        {
            "speaker": 2,
            "line": "So, AGI, huh? Seems like everyone's talking about it these days."
        },
        {
            "speaker": 1,
            "line": "Yeah, it's definitely having a moment, isn't it?"
        },
        {
            "speaker": 2,
            "line": "It is and for good reason, right? I mean, you've been digging into this stuff, listening to the podcasts and everything. What really stood out to you? What got you hooked?"
        },
        {
            "speaker": 1,
            "line": "Honestly, it's the sheer scale of what AGI could do. We're talking about potentially reshaping well everything."
        },
        {
            "speaker": 2,
            "line": "No kidding, but let's be real. Sometimes it feels like every other headline is either hyping AGI up as this technological utopia or painting it as our inevitable robot overlords."
        },
        {
            "speaker": 1,
            "line": "It's easy to get lost in the noise, for sure."
        },
        {
            "speaker": 2,
            "line": "Exactly. So how about we try to cut through some of that, shall we?"
        },
        {
            "speaker": 1,
            "line": "Sounds like a plan."
        },
        {
            "speaker": 2,
            "line": "Okay, so first things first, AGI, what is it really? And I don't just mean some dictionary definition, we're talking about something way bigger than just a super smart computer, right?"
        },
        {
            "speaker": 1,
            "line": "Right, it's not just about more processing power or better algorithms, it's about a fundamental shift in how we think about intelligence itself."
        },
        {
            "speaker": 2,
            "line": "So like, instead of programming a machine for a specific task, we're talking about creating something that can learn and adapt like we do."
        },
        {
            "speaker": 1,
            "line": "Exactly, think of it this way: Right now, we've got AI that can beat a grandmaster at chess but ask that same AI to, say, write a poem or compose a symphony. No chance."
        },
        {
            "speaker": 2,
            "line": "Okay, I see. So, AGI is about bridging that gap, creating something that can move between those different realms of knowledge seamlessly."
        },
        {
            "speaker": 1,
            "line": "Precisely. It's about replicating that uniquely human ability to learn something new and apply that knowledge in completely different contexts and that's a tall order, let me tell you."
        },
        {
            "speaker": 2,
            "line": "I bet. I mean, think about how much we still don't even understand about our own brains."
        },
        {
            "speaker": 1,
            "line": "That's exactly it. We're essentially trying to reverse-engineer something we don't fully comprehend."
        },
        {
            "speaker": 2,
            "line": "And how are researchers even approaching that? What are some of the big ideas out there?"
        },
        {
            "speaker": 1,
            "line": "Well, there are a few different schools of thought. One is this idea of neuromorphic computing where they're literally trying to build computer chips that mimic the structure and function of the human brain."
        },
        {
            "speaker": 2,
            "line": "Wow, so like actually replicating the physical architecture of the brain. That's wild."
        },
        {
            "speaker": 1,
            "line": "It's pretty mind-blowing stuff and then you've got folks working on something called whole brain emulation."
        },
        {
            "speaker": 2,
            "line": "Okay, and what's that all about?"
        },
        {
            "speaker": 1,
            "line": "The basic idea there is to create a complete digital copy of a human brain down to the last neuron and synapse and run it on a sufficiently powerful computer simulation."
        },
        {
            "speaker": 2,
            "line": "Hold on, a digital copy of an entire brain, that sounds like something straight out of science fiction."
        },
        {
            "speaker": 1,
            "line": "It does, doesn't it? But it gives you an idea of the kind of ambition we're talking about here and the truth is we're still a long way off from truly achieving AGI, no matter which approach you look at."
        },
        {
            "speaker": 2,
            "line": "That makes sense but it's still exciting to think about the possibilities, even if they're a ways off."
        },
        {
            "speaker": 1,
            "line": "Absolutely and those possibilities are what really get people fired up about AGI, right? Yeah."
        },
        {
            "speaker": 2,
            "line": "For sure. In fact, I remember you mentioning something in that podcast about AGI's potential to revolutionize scientific research. Something about supercharging breakthroughs."
        },
        {
            "speaker": 1,
            "line": "Oh, absolutely. Imagine an AI that doesn't just crunch numbers but actually understands scientific data the way a human researcher does. We're talking about potential breakthroughs in everything from medicine and healthcare to material science and climate change."
        },
        {
            "speaker": 2,
            "line": "It's like giving scientists this incredibly powerful new tool to tackle some of the biggest challenges we face."
        },
        {
            "speaker": 1,
            "line": "Exactly, it could be a total game changer."
        },
        {
            "speaker": 2,
            "line": "Okay, but let's be real, every coin has two sides. What about the potential downsides of AGI? Because it can't all be sunshine and roses, right?"
        },
        {
            "speaker": 1,
            "line": "Right, there are definitely valid concerns. Probably the biggest one is the impact on the job market. As AGI gets more sophisticated, there's a real chance it could automate a lot of jobs that are currently done by humans."
        },
        {
            "speaker": 2,
            "line": "So we're not just talking about robots taking over factories but potentially things like, what, legal work, analysis, even creative fields?"
        },
        {
            "speaker": 1,
            "line": "Potentially, yes. And that raises a whole host of questions about what happens to those workers, how we retrain them, how we ensure that the benefits of AGI are shared equitably."
        },
        {
            "speaker": 2,
            "line": "Right, because it's not just about the technology itself, but how we choose to integrate it into society."
        },
        {
            "speaker": 1,
            "line": "Absolutely. We need to be having these conversations now about ethics, about regulation, about how to make sure AGI is developed and deployed responsibly."
        },
        {
            "speaker": 2,
            "line": "So it's less about preventing some kind of sci-fi robot apocalypse and more about making sure we're steering this technology in the right direction from the get-go."
        },
        {
            "speaker": 1,
            "line": "Exactly, AGI has the potential to be incredibly beneficial, but it's not going to magically solve all our problems. It's on us to make sure we're using it for good."
        },
        {
            "speaker": 2,
            "line": "It's like you said earlier, it's about shaping the future of intelligence."
        },
        {
            "speaker": 1,
            "line": "I like that. It really is."
        },
        {
            "speaker": 2,
            "line": "And honestly, that's a responsibility that extends beyond just the researchers and the policymakers."
        },
        {
            "speaker": 1,
            "line": "100%"
        },
        {
            "speaker": 2,
            "line": "So to everyone listening out there I'll leave you with this. As AGI continues to develop, what role do you want to play in shaping its future?"
        },
        {
            "speaker": 1,
            "line": "That's a question worth pondering."
        },
        {
            "speaker": 2,
            "line": "It certainly is and on that note, we'll wrap up this deep dive. Thanks for listening, everyone."
        },
        {
            "speaker": 1,
            "line": "Peace."
        }
    ]
}
        """
        system_prompt = f"""
You are a professional podcast generator. Your task is to generate a professional podcast script based on the user input. The user input can also be text extracted from a document.
- The podcast should have 2 speakers.
- The podcast should be long.
- The speakers must not mention each other by name.
- The podcast should be interesting and engaging, and hook the listener from the start.
- The script must be in JSON format.
Follow this example structure:
{example}
"""
        user_prompt = f"Please generate a podcast script based on the following user input:\n{prompt}"

        messages = [
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": user_prompt}
        ]

        response = await self.groq_client.chat.completions.create(
            messages=messages,
            model="llama-3.1-70b-versatile",
            response_format={"type": "json_object"},
            max_tokens=4096,
            temperature=1,
        )

        return json.loads(response.choices[0].message.content)

    async def tts_generate(self, text: str, speaker: int) -> str:
        voice = "en-US-AndrewMultilingualNeural" if speaker == 1 else "en-US-AvaMultilingualNeural"
        speech = edge_tts.Communicate(text, voice)
        
        temp_filename = f"temp_{uuid.uuid4()}.wav"
        try:
            await speech.save(temp_filename)
            return temp_filename
        except Exception as e:
            if os.path.exists(temp_filename):
                os.remove(temp_filename)
            raise e

    async def combine_audio_files(self, audio_files: List[str]) -> str:
        combined_audio = AudioSegment.empty()
        for audio_file in audio_files:
            combined_audio += AudioSegment.from_file(audio_file)
            os.remove(audio_file)  # Clean up temporary files

        output_filename = f"output_{uuid.uuid4()}.wav"
        combined_audio.export(output_filename, format="wav")
        return output_filename

    async def generate_podcast(self, input_text: str) -> str:
        podcast_json = await self.generate_script(input_text)
        print(f"Generated podcast script:\n{podcast_json}")
        audio_files = await asyncio.gather(*[self.tts_generate(item['line'], item['speaker']) for item in podcast_json['podcast']])
        combined_audio = await self.combine_audio_files(audio_files)
        return combined_audio
    
class TextExtractor:
    @staticmethod
    async def extract_from_pdf(file_path: str) -> str:
        async with aiofiles.open(file_path, 'rb') as file:
            content = await file.read()
            pdf_reader = PyPDF2.PdfReader(io.BytesIO(content))
            return " ".join(page.extract_text() for page in pdf_reader.pages)

    @staticmethod
    async def extract_from_txt(file_path: str) -> str:
        async with aiofiles.open(file_path, 'r') as file:
            return await file.read()

    @classmethod
    async def extract_text(cls, file_path: str) -> str:
        _, file_extension = os.path.splitext(file_path)
        if file_extension.lower() == '.pdf':
            return await cls.extract_from_pdf(file_path)
        elif file_extension.lower() == '.txt':
            return await cls.extract_from_txt(file_path)
        else:
            raise ValueError(f"Unsupported file type: {file_extension}")

async def process_input(input_text: str, input_file) -> str:
    if input_file:
        input_text = await TextExtractor.extract_text(input_file.name)

    podcast_generator = PodcastGenerator(groq_api_key=os.environ["GROQ_API_KEY"])
    return await podcast_generator.generate_podcast(input_text)

# Define Gradio interface
iface = gr.Interface(
    fn=process_input,
    inputs=[
        gr.Textbox(label="Input Text"),
        gr.File(label="Or Upload a PDF or TXT file")
    ],
    outputs=[
        gr.Audio(label="Generated Podcast Audio")
    ],
    title="PodcastGen 🎙️",
    description="Generate a 2-speaker podcast from text input or documents!",
    theme="saq1b/gradio-theme"
)

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