Update src/main.py

src/main.py

@@ -18,6 +18,8 @@ from pedalboard import Pedalboard, Reverb, Compressor, HighpassFilter
 from pedalboard.io import AudioFile
 from pydub import AudioSegment
 
+import pyloudnorm as pyln
+
 from mdx import run_mdx
 from rvc import Config, load_hubert, get_vc, rvc_infer
 
@@ -135,6 +137,27 @@ def convert_to_stereo(audio_path):
     else:
         return audio_path
 
+def normalize_audio(input_audio_path, output_audio_path):
+    y1, sr1 = librosa.load(inpud_audio_path, mono = False)
+    meter = pyln.Meter(sr1)
+    lufs_left1 = meter.integrated_loudness(y1[0])
+    lufs_right1 = meter.integrated_loudness(y1[1])
+
+    y2, sr2 = librosa.load(output_audio_path, mono = False)
+    meter2 = pyln.Meter(sr2)
+    lufs_left2 = meter2.integrated_loudness(y2[0])
+    lufs_right2 = meter2.integrated_loudness(y2[1])
+
+    y2_normalized_left = pyln.normalize.loudness(y2[0], lufs_left2, lufs_left1)
+    y2_normalized_right = pyln.normalize.loudness(y2[0], lufs_right2, lufs_right1)
+
+    y2_normalized = np.stack((y2_normalized_left, y2_normalized_right), axis=0)
+    
+    output_path = f"{output_audio_path}_normalized.wav"
+
+    sf.write(output_path, y2_normalized.T, sr2)
+
+    return output_path
 
 def pitch_shift(audio_path, pitch_change):
     output_path = f'{os.path.splitext(audio_path)[0]}_p{pitch_change}.wav'
@@ -290,13 +313,15 @@ def song_cover_pipeline(main_vocals, backup_vocals, voice_model, pitch_change, k
         if backup_vocals == "":
             print("####### ENTRATO QUA, BACKUP VOCALS NON DATI IN INPUT ########")
             ai_vocals_mixed_path = convert_to_stereo(ai_vocals_mixed_path)
+            ai_vocals_mixed_path = normalize(main_vocals, ai_vocals_mixed_path)
             return ai_vocals_mixed_path
             
         display_progress('[~] Combining AI Vocals and Instrumentals...', 0.9, is_webui, progress)
         combine_audio([ai_vocals_mixed_path, backup_vocals], ai_cover_path, main_gain, backup_gain, inst_gain, output_format)
 
         ai_cover_path = convert_to_stereo(ai_cover_path)
-        
+        ai_cover_path = normalize(main_vocals, ai_cover_path)
+    
         return ai_cover_path
 
     except Exception as e: