Update app.py

app.py

@@ -1,139 +1,600 @@
 import gradio as gr
 from pydub import AudioSegment
+from google import genai
+from google.genai import types
+import json
+import uuid
 import edge_tts
 import asyncio
 import aiofiles
 import os
 import time
 import mimetypes
-import uuid
-import json
 from typing import List, Dict
 
-# Model imports
-import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM, StoppingCriteria, StoppingCriteriaList
-
-# Configuration
-MODEL_ID = os.getenv("GEMMA_MODEL_PATH", "tabularisai/german-gemma-3-1b-it")
-HF_TOKEN = os.getenv("Tokentest")  # Optional
-
-# Load tokenizer and model using external snippet
-print(f"Loading model {MODEL_ID}...")
-tokenizer = AutoTokenizer.from_pretrained(
-    MODEL_ID,
-    trust_remote_code=True,
-    use_auth_token=HF_TOKEN
-)
-model = AutoModelForCausalLM.from_pretrained(
-    MODEL_ID,
-    trust_remote_code=True,
-    use_auth_token=HF_TOKEN,
-    torch_dtype=(torch.bfloat16 if torch.cuda.is_available() else torch.float32),
-    device_map="auto"
-).eval()
-
-# Stopping criteria tokens
-PAD = tokenizer.pad_token_id or tokenizer.eos_token_id
-EOT = tokenizer.convert_tokens_to_ids('<end_of_turn>')
+# Constants
+MAX_FILE_SIZE_MB = 20
+MAX_FILE_SIZE_BYTES = MAX_FILE_SIZE_MB * 1024 * 1024  # Convert MB to bytes
 
 class PodcastGenerator:
-    MAX_FILE_MB = 20
-    MAX_FILE_BYTES = MAX_FILE_MB * 1024 * 1024
-
     def __init__(self):
         pass
 
-    async def generate_script(self, prompt: str, language: str, file_obj=None, progress=None) -> Dict:
-        example = '{"topic": "AGI", "podcast": [ ... ] }'
-        lang_inst = (
-            "- The podcast MUST be in the same language as the user input."
-            if language == "Auto Detect"
-            else f"- The podcast MUST be in {language} language"
-        )
-        system_prompt = (
-            "You are a professional podcast generator. Your task is to generate a professional podcast script..."
-            f"\n{lang_inst}\n- The podcast should have 2 speakers.\n- The podcast should be long."
-            "\n- Do not use names for the speakers.\n- The podcast should be interesting, lively, and engaging..."
-            "\n- The script must be in JSON format. Follow this example structure:" + example
-        )
+    async def generate_script(self, prompt: str, language: str, api_key: str, file_obj=None, progress=None) -> 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."
+        }
+    ]
+}
+        """
+
+        if language == "Auto Detect":
+            language_instruction = "- The podcast MUST be in the same language as the user input."
+        else:
+            language_instruction = f"- The podcast MUST be in {language} language"
+
+        system_prompt = f"""
+You are a professional podcast generator. Your task is to generate a professional podcast script based on the user input.
+{language_instruction}
+- The podcast should have 2 speakers.
+- The podcast should be long.
+- Do not use names for the speakers.
+- The podcast should be interesting, lively, and engaging, and hook the listener from the start.
+- The input text might be disorganized or unformatted, originating from sources like PDFs or text files. Ignore any formatting inconsistencies or irrelevant details; your task is to distill the essential points, identify key definitions, and highlight intriguing facts that would be suitable for discussion in a podcast.
+- The script must be in JSON format.
+Follow this example structure:
+{example}
+"""
+        user_prompt = ""
         if prompt and file_obj:
-            user_prompt = f"Generate podcast script based on file and prompt:\n{prompt}"
+            user_prompt = f"Please generate a podcast script based on the uploaded file following user input:\n{prompt}"
         elif prompt:
-            user_prompt = f"Generate podcast script based on prompt:\n{prompt}"
+            user_prompt = f"Please generate a podcast script based on the following user input:\n{prompt}"
         else:
-            user_prompt = "Generate podcast script based on uploaded file."
-
-        full_prompt = system_prompt + "\n\n" + user_prompt
-
-        # sync generation in executor using model.generate
-        def gen_sync():
-            inputs = tokenizer(full_prompt, return_tensors="pt").to(model.device)
-            stop_crit = StoppingCriteriaList([StoppingCriteria(max_length=512)])
-            outputs = model.generate(
-                **inputs,
-                max_new_tokens=512,
-                do_sample=True,
-                pad_token_id=PAD,
-                eos_token_id=EOT
+            user_prompt = "Please generate a podcast script based on the uploaded file."
+
+        messages = []
+        
+        # If file is provided, add it to the messages
+        if file_obj:
+            file_data = await self._read_file_bytes(file_obj)
+            mime_type = self._get_mime_type(file_obj.name)
+            
+            messages.append(
+                types.Content(
+                    role="user",
+                    parts=[
+                        types.Part.from_bytes(
+                            data=file_data,
+                            mime_type=mime_type,
+                        )
+                    ],
+                )
             )
-            return tokenizer.decode(outputs[0], skip_special_tokens=True)
+        
+        # Add text prompt
+        messages.append(
+            types.Content(
+                role="user",
+                parts=[
+                    types.Part.from_text(text=user_prompt)
+                ],
+            )
+        )
 
-        loop = asyncio.get_event_loop()
-        text = await loop.run_in_executor(None, gen_sync)
-        return json.loads(text)
+        client = genai.Client(api_key=api_key)
 
+        safety_settings = [
+            {
+                "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
+                "threshold": "BLOCK_NONE"
+            },
+            {
+                "category": "HARM_CATEGORY_HARASSMENT",
+                "threshold": "BLOCK_NONE"
+            },
+            {
+                "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
+                "threshold": "BLOCK_NONE"
+            },
+            {
+                "category": "HARM_CATEGORY_HATE_SPEECH",
+                "threshold": "BLOCK_NONE"
+            }
+        ]
+
+        try:
+            if progress:
+                progress(0.3, "Generating podcast script...")
+                
+            # Add timeout to the API call
+            response = await asyncio.wait_for(
+                client.aio.models.generate_content(
+                    model="gemini-2.0-flash",
+                    contents=messages,
+                    config=types.GenerateContentConfig(
+                        temperature=1,
+                        response_mime_type="application/json",
+                        safety_settings=[
+                            types.SafetySetting(
+                                category=safety_setting["category"],
+                                threshold=safety_setting["threshold"]
+                            ) for safety_setting in safety_settings
+                        ],
+                        system_instruction=system_prompt
+                    )
+                ),
+                timeout=60  # 60 seconds timeout
+            )
+        except asyncio.TimeoutError:
+            raise Exception("The script generation request timed out. Please try again later.")
+        except Exception as e:
+            if "API key not valid" in str(e):
+                raise Exception("Invalid API key. Please provide a valid Gemini API key.")
+            elif "rate limit" in str(e).lower():
+                raise Exception("Rate limit exceeded for the API key. Please try again later or provide your own Gemini API key.")
+            else:
+                raise Exception(f"Failed to generate podcast script: {e}")
+
+        print(f"Generated podcast script:\n{response.text}")
+        
+        if progress:
+            progress(0.4, "Script generated successfully!")
+            
+        return json.loads(response.text)
+    
     async def _read_file_bytes(self, file_obj) -> bytes:
-        size = getattr(file_obj, 'size', os.path.getsize(file_obj.name))
-        if size > self.MAX_FILE_BYTES:
-            raise Exception(f"File > {self.MAX_FILE_MB}MB")
-        return file_obj.read() if hasattr(file_obj, 'read') else await aiofiles.open(file_obj.name, 'rb').read()
+        """Read file bytes from a file object"""
+        # Check file size before reading
+        if hasattr(file_obj, 'size'):
+            file_size = file_obj.size
+        else:
+            file_size = os.path.getsize(file_obj.name)
+            
+        if file_size > MAX_FILE_SIZE_BYTES:
+            raise Exception(f"File size exceeds the {MAX_FILE_SIZE_MB}MB limit. Please upload a smaller file.")
+            
+        if hasattr(file_obj, 'read'):
+            return file_obj.read()
+        else:
+            async with aiofiles.open(file_obj.name, 'rb') as f:
+                return await f.read()
+    
+    def _get_mime_type(self, filename: str) -> str:
+        """Determine MIME type based on file extension"""
+        ext = os.path.splitext(filename)[1].lower()
+        if ext == '.pdf':
+            return "application/pdf"
+        elif ext == '.txt':
+            return "text/plain"
+        else:
+            # Fallback to the default mime type detector
+            mime_type, _ = mimetypes.guess_type(filename)
+            return mime_type or "application/octet-stream"
 
-    async def tts_generate(self, text: str, speaker: int, s1: str, s2: str) -> str:
-        voice = s1 if speaker == 1 else s2
+    async def tts_generate(self, text: str, speaker: int, speaker1: str, speaker2: str) -> str:
+        voice = speaker1 if speaker == 1 else speaker2
         speech = edge_tts.Communicate(text, voice)
-        fname = f"tmp_{uuid.uuid4()}.wav"
-        await speech.save(fname)
-        return fname
-
-    async def combine_audio_files(self, files: List[str], progress=None) -> str:
-        combined = AudioSegment.empty()
-        for f in files:
-            combined += AudioSegment.from_file(f)
-            os.remove(f)
-        out = f"out_{uuid.uuid4()}.wav"
-        combined.export(out, format="wav")
-        return out
-
-    async def generate_podcast(self, text: str, lang: str, sp1: str, sp2: str, file_obj=None, progress=None) -> str:
-        pj = await self.generate_script(text, lang, file_obj, progress)
-        parts = pj['podcast']
+        
+        temp_filename = f"temp_{uuid.uuid4()}.wav"
+        try:
+            # Add timeout to TTS generation
+            await asyncio.wait_for(speech.save(temp_filename), timeout=30)  # 30 seconds timeout
+            return temp_filename
+        except asyncio.TimeoutError:
+            if os.path.exists(temp_filename):
+                os.remove(temp_filename)
+            raise Exception("Text-to-speech generation timed out. Please try with a shorter text.")
+        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], progress=None) -> str:
+        if progress:
+            progress(0.9, "Combining audio files...")
+            
+        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")
+        
+        if progress:
+            progress(1.0, "Podcast generated successfully!")
+            
+        return output_filename
+
+    async def generate_podcast(self, input_text: str, language: str, speaker1: str, speaker2: str, api_key: str, file_obj=None, progress=None) -> str:
+        try:
+            if progress:
+                progress(0.1, "Starting podcast generation...")
+                
+            # Set overall timeout for the entire process
+            return await asyncio.wait_for(
+                self._generate_podcast_internal(input_text, language, speaker1, speaker2, api_key, file_obj, progress),
+                timeout=600  # 10 minutes total timeout
+            )
+        except asyncio.TimeoutError:
+            raise Exception("The podcast generation process timed out. Please try with shorter text or try again later.")
+        except Exception as e:
+            raise Exception(f"Error generating podcast: {str(e)}")
+    
+    async def _generate_podcast_internal(self, input_text: str, language: str, speaker1: str, speaker2: str, api_key: str, file_obj=None, progress=None) -> str:
+        if progress:
+            progress(0.2, "Generating podcast script...")
+            
+        podcast_json = await self.generate_script(input_text, language, api_key, file_obj, progress)
+        
+        if progress:
+            progress(0.5, "Converting text to speech...")
+        
+        # Process TTS in batches for concurrent processing
         audio_files = []
-        for seg in parts:
-            audio_files.append(await self.tts_generate(seg['line'], seg['speaker'], sp1, sp2))
-        return await self.combine_audio_files(audio_files)
+        total_lines = len(podcast_json['podcast'])
+        
+        # Define batch size to control concurrency
+        batch_size = 10  # Adjust based on system resources
+        
+        # Process in batches
+        for batch_start in range(0, total_lines, batch_size):
+            batch_end = min(batch_start + batch_size, total_lines)
+            batch = podcast_json['podcast'][batch_start:batch_end]
+            
+            # Create tasks for concurrent processing
+            tts_tasks = []
+            for item in batch:
+                tts_task = self.tts_generate(item['line'], item['speaker'], speaker1, speaker2)
+                tts_tasks.append(tts_task)
+            
+            try:
+                # Process batch concurrently
+                batch_results = await asyncio.gather(*tts_tasks, return_exceptions=True)
+                
+                # Check for exceptions and handle results
+                for i, result in enumerate(batch_results):
+                    if isinstance(result, Exception):
+                        # Clean up any files already created
+                        for file in audio_files:
+                            if os.path.exists(file):
+                                os.remove(file)
+                        raise Exception(f"Error generating speech: {str(result)}")
+                    else:
+                        audio_files.append(result)
+                        
+                # Update progress
+                if progress:
+                    current_progress = 0.5 + (0.4 * (batch_end / total_lines))
+                    progress(current_progress, f"Processed {batch_end}/{total_lines} speech segments...")
+            
+            except Exception as e:
+                # Clean up any files already created
+                for file in audio_files:
+                    if os.path.exists(file):
+                        os.remove(file)
+                raise Exception(f"Error in batch TTS generation: {str(e)}")
+        
+        combined_audio = await self.combine_audio_files(audio_files, progress)
+        return combined_audio
+
+async def process_input(input_text: str, input_file, language: str, speaker1: str, speaker2: str, api_key: str = "", progress=None) -> str:
+    start_time = time.time()
+
+    voice_names = {
+        "Andrew - English (United States)": "en-US-AndrewMultilingualNeural",
+        "Ava - English (United States)": "en-US-AvaMultilingualNeural",
+        "Brian - English (United States)": "en-US-BrianMultilingualNeural",
+        "Emma - English (United States)": "en-US-EmmaMultilingualNeural",
+        "Florian - German (Germany)": "de-DE-FlorianMultilingualNeural",
+        "Seraphina - German (Germany)": "de-DE-SeraphinaMultilingualNeural",
+        "Remy - French (France)": "fr-FR-RemyMultilingualNeural",
+        "Vivienne - French (France)": "fr-FR-VivienneMultilingualNeural"
+    }
+
+    speaker1 = voice_names[speaker1]
+    speaker2 = voice_names[speaker2]
+
+    try:
+        if progress:
+            progress(0.05, "Processing input...")
+
+        if not api_key:
+            api_key = os.getenv("GENAI_API_KEY")
+            if not api_key:
+                raise Exception("No API key provided. Please provide a Gemini API key.")
+
+        podcast_generator = PodcastGenerator()
+        podcast = await podcast_generator.generate_podcast(input_text, language, speaker1, speaker2, api_key, input_file, progress)
+
+        end_time = time.time()
+        print(f"Total podcast generation time: {end_time - start_time:.2f} seconds")
+        return podcast
+        
+    except Exception as e:
+        # Ensure we show a user-friendly error
+        error_msg = str(e)
+        if "rate limit" in error_msg.lower():
+            raise Exception("Rate limit exceeded. Please try again later or use your own API key.")
+        elif "timeout" in error_msg.lower():
+            raise Exception("The request timed out. This could be due to server load or the length of your input. Please try again with shorter text.")
+        else:
+            raise Exception(f"Error: {error_msg}")
 
 # Gradio UI
+def generate_podcast_gradio(input_text, input_file, language, speaker1, speaker2, api_key, progress=gr.Progress()):
+    # Handle the file if uploaded
+    file_obj = None
+    if input_file is not None:
+        file_obj = input_file
+        
+    # Use the progress function from Gradio
+    def progress_callback(value, text):
+        progress(value, text)
+
+    # Run the async function in the event loop
+    result = asyncio.run(process_input(
+        input_text, 
+        file_obj, 
+        language, 
+        speaker1, 
+        speaker2, 
+        api_key,
+        progress_callback
+    ))
+    
+    return result
+
+def main():
+    # Define language options
+    language_options = [
+        "Auto Detect",
+        "Afrikaans", "Albanian", "Amharic", "Arabic", "Armenian", "Azerbaijani",
+        "Bahasa Indonesian", "Bangla", "Basque", "Bengali", "Bosnian", "Bulgarian",
+        "Burmese", "Catalan", "Chinese Cantonese", "Chinese Mandarin",
+        "Chinese Taiwanese", "Croatian", "Czech", "Danish", "Dutch", "English",
+        "Estonian", "Filipino", "Finnish", "French", "Galician", "Georgian",
+        "German", "Greek", "Hebrew", "Hindi", "Hungarian", "Icelandic", "Irish",
+        "Italian", "Japanese", "Javanese", "Kannada", "Kazakh", "Khmer", "Korean",
+        "Lao", "Latvian", "Lithuanian", "Macedonian", "Malay", "Malayalam",
+        "Maltese", "Mongolian", "Nepali", "Norwegian Bokmål", "Pashto", "Persian",
+        "Polish", "Portuguese", "Romanian", "Russian", "Serbian", "Sinhala",
+        "Slovak", "Slovene", "Somali", "Spanish", "Sundanese", "Swahili",
+        "Swedish", "Tamil", "Telugu", "Thai", "Turkish", "Ukrainian", "Urdu",
+        "Uzbek", "Vietnamese", "Welsh", "Zulu"
+    ]
+    
+    # Define voice options
+    voice_options = [
+        "Andrew - English (United States)",
+        "Ava - English (United States)",
+        "Brian - English (United States)",
+        "Emma - English (United States)",
+        "Florian - German (Germany)",
+        "Seraphina - German (Germany)",
+        "Remy - French (France)",
+        "Vivienne - French (France)"
+    ]
+    
+    # Create Gradio interface
+    with gr.Blocks(title="PodcastGen 🎙️") as demo:
+        gr.Markdown("# PodcastGen 🎙️")
+        gr.Markdown("Generate a 2-speaker podcast from text input or documents!")
+        
+        with gr.Row():
+            with gr.Column(scale=2):
+                input_text = gr.Textbox(label="Input Text", lines=10, placeholder="Enter text for podcast generation...")
+            
+            with gr.Column(scale=1):
+                input_file = gr.File(label="Or Upload a PDF or TXT file", file_types=[".pdf", ".txt"])
+        
+        with gr.Row():
+            with gr.Column():
+                api_key = gr.Textbox(label="Your Gemini API Key (Optional)", placeholder="Enter API key here if you're getting rate limited", type="password")
+                language = gr.Dropdown(label="Language", choices=language_options, value="Auto Detect")
 
-def run_app():
-    langs = ["Auto Detect","German","English","French"]
-    voices = ["Florian - German (Germany)", "Andrew - English (US)"]
-    gen = PodcastGenerator()
-
-    with gr.Blocks() as demo:
-        inp = gr.Textbox(label="Input Text")
-        file_u = gr.File(label="Upload PDF/TXT")
-        lang_dd = gr.Dropdown(langs, value="Auto Detect", label="Language")
-        sp1 = gr.Dropdown(voices, value=voices[0], label="Speaker 1")
-        sp2 = gr.Dropdown(voices, value=voices[1], label="Speaker 2")
-        out = gr.Audio(label="Podcast", type="filepath")
-        btn = gr.Button("Generate")
-        btn.click(
-            lambda t,f,l,a,b: asyncio.run(gen.generate_podcast(t,l,a,b,f)),
-            inputs=[inp, file_u, lang_dd, sp1, sp2],
-            outputs=[out]
+            with gr.Column():
+                speaker1 = gr.Dropdown(label="Speaker 1 Voice", choices=voice_options, value="Andrew - English (United States)")
+                speaker2 = gr.Dropdown(label="Speaker 2 Voice", choices=voice_options, value="Ava - English (United States)")
+        
+        generate_btn = gr.Button("Generate Podcast", variant="primary")
+        
+        with gr.Row():
+            output_audio = gr.Audio(label="Generated Podcast", type="filepath", format="wav")
+            
+        generate_btn.click(
+            fn=generate_podcast_gradio,
+            inputs=[input_text, input_file, language, speaker1, speaker2, api_key],
+            outputs=[output_audio]
         )
+    
     demo.launch()
 
-if __name__ == '__main__':
-    run_app()
+if __name__ == "__main__":
+    main()