Update app.py

app.py

@@ -43,14 +43,9 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
         combined_prompt = f"{system_prompt}\nUser concept: {user_prompt}\nRefined script and music suggestion:"
         result = llama_pipeline(combined_prompt, max_new_tokens=200, do_sample=True, temperature=0.9)
 
-        generated_text = result[0]["generated_text"]
-
-        # Check if the delimiter exists
-        if "Music Suggestion:" in generated_text:
-            script, music_suggestion = generated_text.split("Music Suggestion:", 1)
-            return script.strip(), music_suggestion.strip()
-        else:
-            return "Error: 'Music Suggestion:' delimiter not found in the generated text.", None
+        generated_text = result[0]["generated_text"].split("Refined script and music suggestion:")[-1].strip()
+        script, music_suggestion = generated_text.split("Music Suggestion:")
+        return script.strip(), music_suggestion.strip()
     except Exception as e:
         return f"Error generating script: {e}", None
 
@@ -60,12 +55,16 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
 @spaces.GPU(duration=300)
 def generate_voice(script: str, speaker: str):
     try:
-        from TTS.api import TTS
+        # Replace with your chosen TTS model
+        tts_model = "coqui/XTTS-v2"
+        processor = AutoProcessor.from_pretrained(tts_model)
+        model = AutoModelForCausalLM.from_pretrained(tts_model)
 
-        tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2", gpu=True)
+        inputs = processor(script, return_tensors="pt")
+        speech = model.generate(**inputs)
 
         output_path = f"{tempfile.gettempdir()}/generated_voice.wav"
-        tts.tts_to_file(text=script, file_path=output_path)
+        write(output_path, 22050, speech.cpu().numpy())
         return output_path
     except Exception as e:
         return f"Error generating voice-over: {e}"
@@ -141,8 +140,8 @@ with gr.Blocks() as demo:
     final_output = gr.Audio(label="Final Promo Audio", type="filepath")
 
     generate_script_button.click(
-        fn=generate_script,
-        inputs=[user_prompt, llama_model_id, hf_token, duration],
+        fn=lambda user_prompt, llama_model_id, duration: generate_script(user_prompt, llama_model_id, hf_token, duration),
+        inputs=[user_prompt, llama_model_id, duration],
         outputs=[script_output],
     )