Update app.py

app.py

@@ -50,6 +50,7 @@ def get_llama_pipeline(model_id: str, token: str):
     """
     if model_id in LLAMA_PIPELINES:
         return LLAMA_PIPELINES[model_id]
+
     tokenizer = AutoTokenizer.from_pretrained(model_id, use_auth_token=token)
     model = AutoModelForCausalLM.from_pretrained(
         model_id,
@@ -62,6 +63,7 @@ def get_llama_pipeline(model_id: str, token: str):
     LLAMA_PIPELINES[model_id] = text_pipeline
     return text_pipeline
 
+
 def get_musicgen_model(model_key: str = "facebook/musicgen-large"):
     """
     Returns a cached MusicGen model if available; otherwise, loads it.
@@ -69,6 +71,7 @@ def get_musicgen_model(model_key: str = "facebook/musicgen-large"):
     """
     if model_key in MUSICGEN_MODELS:
         return MUSICGEN_MODELS[model_key]
+
     model = MusicgenForConditionalGeneration.from_pretrained(model_key)
     processor = AutoProcessor.from_pretrained(model_key)
     device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -76,16 +79,19 @@ def get_musicgen_model(model_key: str = "facebook/musicgen-large"):
     MUSICGEN_MODELS[model_key] = (model, processor)
     return model, processor
 
+
 def get_tts_model(model_name: str = "tts_models/en/ljspeech/tacotron2-DDC"):
     """
     Returns a cached TTS model if available; otherwise, loads it.
     """
     if model_name in TTS_MODELS:
         return TTS_MODELS[model_name]
+
     tts_model = TTS(model_name)
     TTS_MODELS[model_name] = tts_model
     return tts_model
 
+
 # ---------------------------------------------------------------------
 # Script Generation Function
 # ---------------------------------------------------------------------
@@ -97,6 +103,7 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
     """
     try:
         text_pipeline = get_llama_pipeline(model_id, token)
+
         system_prompt = (
             "You are an expert radio imaging producer specializing in sound design and music. "
             f"Based on the user's concept and the selected duration of {duration} seconds, produce the following: "
@@ -105,6 +112,7 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
             "3. Music styles or track recommendations. Prefix this section with 'Music Suggestions:'."
         )
         combined_prompt = f"{system_prompt}\nUser concept: {user_prompt}\nOutput:"
+
         with torch.inference_mode():
             result = text_pipeline(
                 combined_prompt,
@@ -112,14 +120,17 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
                 do_sample=True,
                 temperature=0.8
             )
+
         generated_text = result[0]["generated_text"]
         if "Output:" in generated_text:
             generated_text = generated_text.split("Output:")[-1].strip()
+
         # Default placeholders
         voice_script = "No voice-over script found."
         sound_design = "No sound design suggestions found."
         music_suggestions = "No music suggestions found."
-        # Extract sections if present
+
+        # Voice-Over Script
         if "Voice-Over Script:" in generated_text:
             parts = generated_text.split("Voice-Over Script:")
             voice_script_part = parts[1]
@@ -127,6 +138,8 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
                 voice_script = voice_script_part.split("Sound Design Suggestions:")[0].strip()
             else:
                 voice_script = voice_script_part.strip()
+
+        # Sound Design
         if "Sound Design Suggestions:" in generated_text:
             parts = generated_text.split("Sound Design Suggestions:")
             sound_design_part = parts[1]
@@ -134,43 +147,17 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
                 sound_design = sound_design_part.split("Music Suggestions:")[0].strip()
             else:
                 sound_design = sound_design_part.strip()
+
+        # Music Suggestions
         if "Music Suggestions:" in generated_text:
             parts = generated_text.split("Music Suggestions:")
             music_suggestions = parts[1].strip()
+
         return voice_script, sound_design, music_suggestions
+
     except Exception as e:
         return f"Error generating script: {e}", "", ""
 
-# ---------------------------------------------------------------------
-# Ad Promo Idea Generation Function
-# ---------------------------------------------------------------------
-@spaces.GPU(duration=100)
-def generate_ad_promo_idea(user_prompt: str, model_id: str, token: str):
-    """
-    Generates a creative ad promo idea based on the user's concept.
-    Returns a string containing the ad promo idea.
-    """
-    try:
-        text_pipeline = get_llama_pipeline(model_id, token)
-        system_prompt = (
-            "You are a creative advertising strategist. "
-            "Generate a unique and engaging ad promo idea based on the following concept. "
-            "Include creative angles, potential taglines, and media suggestions."
-        )
-        combined_prompt = f"{system_prompt}\nConcept: {user_prompt}\nAd Promo Idea:"
-        with torch.inference_mode():
-            result = text_pipeline(
-                combined_prompt,
-                max_new_tokens=150,
-                do_sample=True,
-                temperature=0.8
-            )
-        generated_text = result[0]["generated_text"]
-        if "Ad Promo Idea:" in generated_text:
-            generated_text = generated_text.split("Ad Promo Idea:")[-1].strip()
-        return generated_text
-    except Exception as e:
-        return f"Error generating ad promo idea: {e}"
 
 # ---------------------------------------------------------------------
 # Voice-Over Generation Function
@@ -184,14 +171,21 @@ def generate_voice(script: str, tts_model_name: str = "tts_models/en/ljspeech/ta
     try:
         if not script.strip():
             return "Error: No script provided."
+
+        # Clean the script to remove special characters (e.g., asterisks) that may produce warnings
         cleaned_script = clean_text(script)
+
         tts_model = get_tts_model(tts_model_name)
+
+        # Generate and save voice
         output_path = os.path.join(tempfile.gettempdir(), "voice_over.wav")
         tts_model.tts_to_file(text=cleaned_script, file_path=output_path)
         return output_path
+
     except Exception as e:
         return f"Error generating voice: {e}"
 
+
 # ---------------------------------------------------------------------
 # Music Generation Function
 # ---------------------------------------------------------------------
@@ -204,23 +198,30 @@ def generate_music(prompt: str, audio_length: int):
     try:
         if not prompt.strip():
             return "Error: No music suggestion provided."
+
         model_key = "facebook/musicgen-large"
         musicgen_model, musicgen_processor = get_musicgen_model(model_key)
+
         device = "cuda" if torch.cuda.is_available() else "cpu"
         # Process the input and move each tensor to the proper device
         inputs = musicgen_processor(text=[prompt], padding=True, return_tensors="pt")
         inputs = {k: v.to(device) for k, v in inputs.items()}
+
         with torch.inference_mode():
             outputs = musicgen_model.generate(**inputs, max_new_tokens=audio_length)
-        # Post-process the output to create a WAV file
+
         audio_data = outputs[0, 0].cpu().numpy()
         normalized_audio = (audio_data / max(abs(audio_data)) * 32767).astype("int16")
+
         output_path = os.path.join(tempfile.gettempdir(), "musicgen_large_generated_music.wav")
         write(output_path, 44100, normalized_audio)
+
         return output_path
+
     except Exception as e:
         return f"Error generating music: {e}"
 
+
 # ---------------------------------------------------------------------
 # Audio Blending with Duration Sync & Ducking
 # ---------------------------------------------------------------------
@@ -228,33 +229,46 @@ def generate_music(prompt: str, audio_length: int):
 def blend_audio(voice_path: str, music_path: str, ducking: bool, duck_level: int = 10):
     """
     Blends two audio files (voice and music).
+    1. If music < voice, loops the music until it meets/exceeds the voice duration.
+    2. If music > voice, trims music to the voice duration.
+    3. If ducking=True, the music is attenuated by 'duck_level' dB while the voice is playing.
     Returns the file path to the blended .wav file.
     """
     try:
         if not os.path.isfile(voice_path) or not os.path.isfile(music_path):
             return "Error: Missing audio files for blending."
+
         voice = AudioSegment.from_wav(voice_path)
         music = AudioSegment.from_wav(music_path)
-        voice_len = len(voice)
-        music_len = len(music)
+
+        voice_len = len(voice)  # in milliseconds
+        music_len = len(music)  # in milliseconds
+
+        # Loop music if it's shorter than the voice
         if music_len < voice_len:
             looped_music = AudioSegment.empty()
             while len(looped_music) < voice_len:
                 looped_music += music
             music = looped_music
+
+        # Trim music if it's longer than the voice
         if len(music) > voice_len:
             music = music[:voice_len]
+
         if ducking:
             ducked_music = music - duck_level
             final_audio = ducked_music.overlay(voice)
         else:
             final_audio = music.overlay(voice)
+
         output_path = os.path.join(tempfile.gettempdir(), "blended_output.wav")
         final_audio.export(output_path, format="wav")
         return output_path
+
     except Exception as e:
         return f"Error blending audio: {e}"
 
+
 # ---------------------------------------------------------------------
 # Gradio Interface with Enhanced UI
 # ---------------------------------------------------------------------
@@ -274,23 +288,19 @@ with gr.Blocks(css="""
     }
     .header h1 {
         margin: 0;
-        font-size: 2.8rem;
+        font-size: 2.5rem;
     }
     .header p {
         font-size: 1.2rem;
     }
-    .instructions {
-        background-color: #2e2e2e;
-        border-radius: 8px;
-        padding: 1rem;
-        margin-bottom: 1rem;
-        font-size: 0.95rem;
-    }
     .gradio-container {
         background: #2e2e2e;
         border-radius: 10px;
         padding: 1rem;
-        margin-bottom: 1rem;
+    }
+    .tab-title {
+        font-size: 1.1rem;
+        font-weight: bold;
     }
     .footer {
         text-align: center;
@@ -299,64 +309,31 @@ with gr.Blocks(css="""
         padding: 1rem;
         color: #cccccc;
     }
-    .btn-clear {
-        margin-left: 1rem;
-        background: #ff5555;
-        color: #fff;
-    }
 """) as demo:
 
     # Custom Header
     with gr.Row(elem_classes="header"):
         gr.Markdown("""
-        <h1>🎧 AI Ads Promo</h1>
-        <p>Your all-in-one AI solution for crafting engaging audio ads. <br><em>Demo MVP</em></p>
+        <h1>🎧 AI Promo Studio</h1>
+        <p>Your all-in-one AI solution for crafting engaging audio promos.</p>
         """)
 
     gr.Markdown("""
-    Welcome to **AI Ads Promo (Demo MVP)**! This platform leverages state-of-the-art AI models to help you generate creative advertising content.
-    Use the tabs below to generate:
-    - **Ad Promo Ideas**
-    - **Voice-Over Scripts**
-    - **Natural-Sounding Voice-Overs**
-    - **Custom Music Tracks**
-    - **Blended Audio Ads**
+    Welcome to **AI Promo Studio**! This platform leverages state-of-the-art AI models to help you generate:
+    
+    - **Script**: Generate a compelling voice-over script with LLaMA.
+    - **Voice Synthesis**: Create natural-sounding voice-overs using Coqui TTS.
+    - **Music Production**: Produce custom music tracks with MusicGen.
+    - **Audio Blending**: Seamlessly blend voice and music with options for ducking.
     """)
 
     with gr.Tabs():
-        # Tab 1: Ad Promo Idea Generation
-        with gr.Tab("💡 Ad Promo Idea"):
-            gr.Markdown("Enter a concept for your ad and let the system generate a creative ad promo idea with taglines and media suggestions.")
-            with gr.Row():
-                ad_concept = gr.Textbox(
-                    label="Ad Concept",
-                    placeholder="E.g., A vibrant summer sale for a trendy clothing brand...",
-                    lines=2
-                )
-            with gr.Row():
-                llama_model_id_idea = gr.Textbox(
-                    label="LLaMA Model ID", 
-                    value="meta-llama/Meta-Llama-3-8B-Instruct", 
-                    placeholder="Enter a valid Hugging Face model ID"
-                )
-            with gr.Row():
-                generate_ad_idea_button = gr.Button("Generate Ad Promo Idea", variant="primary")
-                clear_ad_idea = gr.Button("Clear", variant="stop", elem_classes="btn-clear")
-            ad_idea_output = gr.Textbox(label="Generated Ad Promo Idea", lines=5, interactive=False)
-            generate_ad_idea_button.click(
-                fn=lambda concept, model_id: generate_ad_promo_idea(concept, model_id, HF_TOKEN),
-                inputs=[ad_concept, llama_model_id_idea],
-                outputs=ad_idea_output
-            )
-            clear_ad_idea.click(fn=lambda: "", inputs=None, outputs=ad_idea_output)
-
-        # Tab 2: Script Generation
+        # Step 1: Generate Script
         with gr.Tab("📝 Script Generation"):
-            gr.Markdown("Generate a voice-over script along with sound design and music suggestions based on your promo idea.")
             with gr.Row():
                 user_prompt = gr.Textbox(
                     label="Promo Idea", 
-                    placeholder="E.g., A 30-second energetic promo for a new product launch...",
+                    placeholder="E.g., A 30-second promo for a morning show...",
                     lines=2
                 )
             with gr.Row():
@@ -372,22 +349,20 @@ with gr.Blocks(css="""
                     step=15, 
                     value=30
                 )
-            with gr.Row():
-                generate_script_button = gr.Button("Generate Script", variant="primary")
-                clear_script = gr.Button("Clear", variant="stop", elem_classes="btn-clear")
-            script_output = gr.Textbox(label="Voice-Over Script", lines=5, interactive=False)
+            generate_script_button = gr.Button("Generate Script", variant="primary")
+            script_output = gr.Textbox(label="Generated Voice-Over Script", lines=5, interactive=False)
             sound_design_output = gr.Textbox(label="Sound Design Suggestions", lines=3, interactive=False)
             music_suggestion_output = gr.Textbox(label="Music Suggestions", lines=3, interactive=False)
+
             generate_script_button.click(
                 fn=lambda user_prompt, model_id, dur: generate_script(user_prompt, model_id, HF_TOKEN, dur),
                 inputs=[user_prompt, llama_model_id, duration],
-                outputs=[script_output, sound_design_output, music_suggestion_output]
+                outputs=[script_output, sound_design_output, music_suggestion_output],
             )
-            clear_script.click(fn=lambda: ["", "", ""], inputs=None, outputs=[script_output, sound_design_output, music_suggestion_output])
 
-        # Tab 3: Voice Synthesis
+        # Step 2: Generate Voice
         with gr.Tab("🎤 Voice Synthesis"):
-            gr.Markdown("Convert your generated script into a natural-sounding voice-over using Coqui TTS.")
+            gr.Markdown("Generate a natural-sounding voice-over using Coqui TTS.")
             selected_tts_model = gr.Dropdown(
                 label="TTS Model",
                 choices=[
@@ -398,19 +373,18 @@ with gr.Blocks(css="""
                 value="tts_models/en/ljspeech/tacotron2-DDC",
                 multiselect=False
             )
-            with gr.Row():
-                generate_voice_button = gr.Button("Generate Voice-Over", variant="primary")
-                clear_voice = gr.Button("Clear", variant="stop", elem_classes="btn-clear")
+            generate_voice_button = gr.Button("Generate Voice-Over", variant="primary")
             voice_audio_output = gr.Audio(label="Voice-Over (WAV)", type="filepath")
+
             generate_voice_button.click(
                 fn=lambda script, tts_model: generate_voice(script, tts_model),
-                inputs=script_output, outputs=voice_audio_output
+                inputs=[script_output, selected_tts_model],
+                outputs=voice_audio_output,
             )
-            clear_voice.click(fn=lambda: "", inputs=None, outputs=voice_audio_output)
 
-        # Tab 4: Music Production
+        # Step 3: Generate Music
         with gr.Tab("🎶 Music Production"):
-            gr.Markdown("Generate a custom music track based on the suggestions using the MusicGen model.")
+            gr.Markdown("Generate a custom music track using the **MusicGen Large** model.")
             audio_length = gr.Slider(
                 label="Music Length (tokens)",
                 minimum=128, 
@@ -419,20 +393,18 @@ with gr.Blocks(css="""
                 value=512,
                 info="Increase tokens for longer audio (inference time may vary)."
             )
-            with gr.Row():
-                generate_music_button = gr.Button("Generate Music", variant="primary")
-                clear_music = gr.Button("Clear", variant="stop", elem_classes="btn-clear")
+            generate_music_button = gr.Button("Generate Music", variant="primary")
             music_output = gr.Audio(label="Generated Music (WAV)", type="filepath")
+
             generate_music_button.click(
                 fn=lambda music_suggestion, length: generate_music(music_suggestion, length),
                 inputs=[music_suggestion_output, audio_length],
-                outputs=[music_output]
+                outputs=[music_output],
             )
-            clear_music.click(fn=lambda: "", inputs=None, outputs=music_output)
 
-        # Tab 5: Audio Blending
+        # Step 4: Blend Audio
         with gr.Tab("🎚️ Audio Blending"):
-            gr.Markdown("Blend your voice-over and music track. Music will be adjusted to match the voice duration with an option to enable ducking.")
+            gr.Markdown("Blend your voice-over and music track. Music will be looped/truncated to match the voice duration. Enable ducking to lower the music during voice segments.")
             ducking_checkbox = gr.Checkbox(label="Enable Ducking?", value=True)
             duck_level_slider = gr.Slider(
                 label="Ducking Level (dB attenuation)", 
@@ -441,16 +413,14 @@ with gr.Blocks(css="""
                 step=1, 
                 value=10
             )
-            with gr.Row():
-                blend_button = gr.Button("Blend Voice + Music", variant="primary")
-                clear_blend = gr.Button("Clear", variant="stop", elem_classes="btn-clear")
+            blend_button = gr.Button("Blend Voice + Music", variant="primary")
             blended_output = gr.Audio(label="Final Blended Output (WAV)", type="filepath")
+
             blend_button.click(
                 fn=blend_audio,
                 inputs=[voice_audio_output, music_output, ducking_checkbox, duck_level_slider],
                 outputs=blended_output
             )
-            clear_blend.click(fn=lambda: "", inputs=None, outputs=blended_output)
 
     # Footer
     gr.Markdown("""
@@ -458,10 +428,11 @@ with gr.Blocks(css="""
         <hr>
         Created with ❤️ by <a href="https://bilsimaging.com" target="_blank" style="color: #88aaff;">bilsimaging.com</a>
         <br>
-        <small>AI Ads Promo (Demo MVP) &copy; 2025</small>
+        <small>AI Promo Studio &copy; 2025</small>
     </div>
     """)
     
+    # Visitor Badge
     gr.HTML("""
     <div style="text-align: center; margin-top: 1rem;">
         <a href="https://visitorbadge.io/status?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2FBils%2Fradiogold">
@@ -471,4 +442,3 @@ with gr.Blocks(css="""
     """)
 
 demo.launch(debug=True)
-