Update app.py

app.py

@@ -4,6 +4,8 @@ import torch
 import gradio as gr
 from transformers import AutoModelForCausalLM, AutoTokenizer
 from huggingface_hub import snapshot_download
+import google.generativeai as genai
+import re
 import logging
 
 logging.basicConfig(level=logging.INFO)
@@ -28,6 +30,40 @@ VOICES = ["tara", "leah", "jess", "leo", "dan", "mia", "zac", "zoe"]
 # Available Emotive Tags
 EMOTIVE_TAGS = ["`<laugh>`", "`<chuckle>`", "`<sigh>`", "`<cough>`", "`<sniffle>`", "`<groan>`", "`<yawn>`", "`<gasp>`"]
 
+@spaces.GPU()
+def generate_podcast_script(api_key, prompt, uploaded_file, duration, num_hosts):
+    try:
+        genai.configure(api_key=api_key)
+        model = genai.GenerativeModel('gemini-2.5-pro-preview-03-25')
+        
+        combined_content = prompt or ""
+        if uploaded_file:
+            file_content = uploaded_file.read().decode('utf-8')
+            combined_content += "\n" + file_content if combined_content else file_content
+        
+        prompt = f"""
+        Create a podcast script for {'one person' if num_hosts == 1 else 'two people'} discussing:
+        {combined_content}
+        
+        Duration: {duration} minutes. Include natural speech, humor, and occasional off-topic thoughts.
+        Use speech fillers like um, ah. Vary emotional tone.
+        
+        Format: {'Monologue' if num_hosts == 1 else 'Alternating dialogue'} without speaker labels.
+        Separate {'paragraphs' if num_hosts == 1 else 'lines'} with blank lines.
+        
+        Use emotion tags in angle brackets: <laugh>, <sigh>, <chuckle>, <cough>, <sniffle>, <groan>, <yawn>, <gasp>.
+        
+        Example: "I can't believe I stayed up all night <yawn> only to find out the meeting was canceled <groan>."
+        
+        Ensure content flows naturally and stays on topic. Match the script length to {duration} minutes.
+        """
+        
+        response = model.generate_content(prompt)
+        return re.sub(r'[^a-zA-Z0-9\s.,?!<>]', '', response.text)
+    except Exception as e:
+        logger.error(f"Error generating podcast script: {str(e)}")
+        raise
+
 def process_prompt(prompt, voice, tokenizer, device):
     prompt = f"{voice}: {prompt}"
     input_ids = tokenizer(prompt, return_tensors="pt").input_ids
@@ -122,18 +158,17 @@ def generate_speech(text, voice, temperature, top_p, repetition_penalty, max_new
         return None
 
 with gr.Blocks(title="Orpheus Text-to-Speech") as demo:
-    gr.Markdown(f"""
-    # 🎵 [Orpheus Text-to-Speech](https://github.com/canopyai/Orpheus-TTS)
-    Enter your text below and hear it converted to natural-sounding speech with the Orpheus TTS model.
-    
-    ## Tips for better prompts:
-    - Add paralinguistic elements like {", ".join(EMOTIVE_TAGS)} or `uhm` for more human-like speech.
-    - Longer text prompts generally work better than very short phrases
-    """)
-
     with gr.Row():
+        with gr.Column(scale=1):
+            gemini_api_key = gr.Textbox(label="Gemini API Key", type="password")
+            prompt = gr.Textbox(label="Prompt", lines=8, placeholder="Enter your text here...")
+            uploaded_file = gr.File(label="Upload File")
+            duration = gr.Slider(minimum=1, maximum=60, value=5, step=1, label="Duration (minutes)")
+            num_hosts = gr.Radio(["1", "2"], label="Number of Hosts", value="1")
+            generate_script_btn = gr.Button("Generate Podcast Script")
+
         with gr.Column(scale=2):
-            text_input = gr.Textbox(label="Text to speak", lines=5)
+            script_output = gr.Textbox(label="Generated Script", lines=10)
             voice = gr.Dropdown(
                 choices=VOICES,
                 value="tara",
@@ -170,16 +205,22 @@ with gr.Blocks(title="Orpheus Text-to-Speech") as demo:
         with gr.Column(scale=2):
             audio_output = gr.Audio(label="Generated Speech", type="numpy")
             
+    generate_script_btn.click(
+        fn=generate_podcast_script,
+        inputs=[gemini_api_key, prompt, uploaded_file, duration, num_hosts],
+        outputs=script_output
+    )
+    
     submit_btn.click(
         fn=generate_speech,
-        inputs=[text_input, voice, temperature, top_p, repetition_penalty, max_new_tokens],
+        inputs=[script_output, voice, temperature, top_p, repetition_penalty, max_new_tokens],
         outputs=audio_output
     )
     
     clear_btn.click(
-        fn=lambda: (None, None),
+        fn=lambda: (None, None, None),
         inputs=[],
-        outputs=[text_input, audio_output]
+        outputs=[prompt, script_output, audio_output]
     )
 
 if __name__ == "__main__":