loudness issues

app.py

@@ -197,33 +197,36 @@ def continue_music(input_audio_path, prompt_duration, musicgen_model, output_dur
         generated_audio_segment = AudioSegment.from_wav(filename_with_extension)
         file_paths_for_cleanup.append(filename_with_extension)
 
-    # VOLUME MATCHING: Use RMS instead of peak normalization
+    # SMART VOLUME MATCHING: Only match the prompt portion
 
     # 1. Remove prompt duration from original (no overlap)  
     prompt_duration_ms = int(prompt_duration * 1000)
     original_minus_prompt = original_audio[:-prompt_duration_ms]
 
-    # 2. Calculate RMS levels for perceived loudness
+    # 2. Extract JUST the prompt portion from generated audio for RMS analysis
+    generated_prompt_portion = generated_audio_segment[:prompt_duration_ms]
+
+    # 3. Calculate RMS of the transition points
     original_rms = original_minus_prompt.rms
-    generated_rms = generated_audio_segment.rms
+    prompt_portion_rms = generated_prompt_portion.rms
 
-    print(f"🔊 Volume analysis:")
-    print(f"   Original RMS: {original_rms}")
-    print(f"   Generated RMS: {generated_rms}")
+    print(f"🔊 Smart volume analysis:")
+    print(f"   Original ending RMS: {original_rms}")
+    print(f"   Generated prompt RMS: {prompt_portion_rms}")
+    print(f"   Generated full RMS: {generated_audio_segment.rms}")
 
-    # 3. Match generated segment to original's RMS level
-    if generated_rms > 0:  # Avoid division by zero
-        # Calculate dB adjustment needed
+    # 4. Match the prompt portion to original level
+    if prompt_portion_rms > 0:
         from pydub.utils import ratio_to_db
-        volume_adjustment = ratio_to_db(original_rms / generated_rms)
-        print(f"   Applying {volume_adjustment:.1f}dB to generated audio")
+        volume_adjustment = ratio_to_db(original_rms / prompt_portion_rms)
+        print(f"   Applying {volume_adjustment:.1f}dB to entire generated segment")
         
-        # Apply volume adjustment
+        # Apply to entire segment (preserves the buildup)
         generated_matched = generated_audio_segment + volume_adjustment
     else:
         generated_matched = generated_audio_segment
 
-    # 4. Combine seamlessly
+    # 5. Combine seamlessly
     combined_audio = original_minus_prompt + generated_matched
     
     # Save final result