Update app.py

app.py

@@ -8,7 +8,6 @@ from transformers import (
     AutoProcessor,
     MusicgenForConditionalGeneration,
 )
-from diffusers import StableDiffusionPipeline
 from scipy.io.wavfile import write
 from pydub import AudioSegment
 from dotenv import load_dotenv
@@ -25,6 +24,7 @@ hf_token = os.getenv("HF_TOKEN")
 # Script Generation Function
 # ---------------------------------------------------------------------
 @spaces.GPU(duration=300)
+@spaces.GPU(duration=300)
 def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
     try:
         tokenizer = AutoTokenizer.from_pretrained(model_id, use_auth_token=token)
@@ -87,36 +87,24 @@ def generate_voice(script: str, speaker: str = "default"):
 # Music Generation Function
 # ---------------------------------------------------------------------
 @spaces.GPU(duration=300)
-def generate_music(prompt: str, audio_length: int, model_choice: str):
+def generate_music(prompt: str, audio_length: int):
     try:
-        if model_choice == "Stable Audio Open 1.0":
-            stable_pipeline = StableDiffusionPipeline.from_pretrained("stabilityai/stable-audio-open-1.0")
-            stable_pipeline.to("cuda" if torch.cuda.is_available() else "cpu")
-
-            audio = stable_pipeline(prompt, num_inference_steps=50, guidance_scale=7.5)
-            output_path = f"{tempfile.gettempdir()}/stable_generated_music.wav"
-            write(output_path, 44100, audio["sample"].cpu().numpy())
-            return output_path
+        musicgen_model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
+        musicgen_processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
 
-        elif model_choice == "MusicGen":
-            musicgen_model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
-            musicgen_processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+        musicgen_model.to(device)
 
-            device = "cuda" if torch.cuda.is_available() else "cpu"
-            musicgen_model.to(device)
+        inputs = musicgen_processor(text=[prompt], padding=True, return_tensors="pt").to(device)
+        outputs = musicgen_model.generate(**inputs, max_new_tokens=audio_length)
 
-            inputs = musicgen_processor(text=[prompt], padding=True, return_tensors="pt").to(device)
-            outputs = musicgen_model.generate(**inputs, max_new_tokens=audio_length)
+        audio_data = outputs[0, 0].cpu().numpy()
+        normalized_audio = (audio_data / max(abs(audio_data)) * 32767).astype("int16")
 
-            audio_data = outputs[0, 0].cpu().numpy()
-            normalized_audio = (audio_data / max(abs(audio_data)) * 32767).astype("int16")
+        output_path = f"{tempfile.gettempdir()}/generated_music.wav"
+        write(output_path, 44100, normalized_audio)
 
-            output_path = f"{tempfile.gettempdir()}/musicgen_generated_music.wav"
-            write(output_path, 44100, normalized_audio)
-            return output_path
-
-        else:
-            return "Invalid model choice!"
+        return output_path
     except Exception as e:
         return f"Error generating music: {e}"
 
@@ -144,12 +132,11 @@ def blend_audio(voice_path: str, music_path: str, ducking: bool):
 # ---------------------------------------------------------------------
 with gr.Blocks() as demo:
     gr.Markdown("""
-        # 🎵 AI Promo Studio 🚀
-        Generate scripts, sound design, and music suggestions with ease.
+        # 🎧 AI Promo Studio with Step-by-Step Script, Voice, Music, and Mixing 🚀  
+        Generate and mix radio promos effortlessly with AI tools!
     """)
 
     with gr.Tabs():
-        # Step 1: Generate Script
         with gr.Tab("Step 1: Generate Script"):
             with gr.Row():
                 user_prompt = gr.Textbox(label="Promo Idea", placeholder="E.g., A 30-second promo for a morning show.")
@@ -157,17 +144,15 @@ with gr.Blocks() as demo:
                 duration = gr.Slider(label="Duration (seconds)", minimum=15, maximum=60, step=15, value=30)
 
             generate_script_button = gr.Button("Generate Script")
-            script_output = gr.Textbox(label="Generated Voice-Over Script", lines=5)
-            sound_design_output = gr.Textbox(label="Sound Design Ideas", lines=3)
-            music_suggestion_output = gr.Textbox(label="Music Suggestions", lines=3)
+            script_output = gr.Textbox(label="Generated Script")
+            music_suggestion_output = gr.Textbox(label="Music Suggestion")
 
             generate_script_button.click(
                 fn=lambda user_prompt, model_id, duration: generate_script(user_prompt, model_id, hf_token, duration),
                 inputs=[user_prompt, llama_model_id, duration],
-                outputs=[script_output, sound_design_output, music_suggestion_output],
+                outputs=[script_output, music_suggestion_output],
             )
 
-        # Step 2: Generate Voice
         with gr.Tab("Step 2: Generate Voice"):
             with gr.Row():
                 speaker = gr.Textbox(label="Voice Style (optional)", placeholder="E.g., male, female, or neutral.")
@@ -181,22 +166,16 @@ with gr.Blocks() as demo:
                 outputs=[voice_output],
             )
 
-        # Step 3: Generate Music
         with gr.Tab("Step 3: Generate Music"):
             with gr.Row():
                 audio_length = gr.Slider(label="Music Length (tokens)", minimum=128, maximum=1024, step=64, value=512)
-                model_choice = gr.Dropdown(
-                    label="Select Music Generation Model",
-                    choices=["Stable Audio Open 1.0", "MusicGen"],
-                    value="Stable Audio Open 1.0"
-                )
 
             generate_music_button = gr.Button("Generate Music")
             music_output = gr.Audio(label="Generated Music", type="filepath")
 
             generate_music_button.click(
-                fn=lambda music_suggestion, audio_length, model_choice: generate_music(music_suggestion, audio_length, model_choice),
-                inputs=[music_suggestion_output, audio_length, model_choice],
+                fn=lambda music_suggestion, audio_length: generate_music(music_suggestion, audio_length),
+                inputs=[music_suggestion_output, audio_length],
                 outputs=[music_output],
             )
 
@@ -220,4 +199,4 @@ with gr.Blocks() as demo:
         </p>
     """)
 
-demo.launch(debug=True)
+demo.launch(debug=True)