changed back to [0,1] normal

data.py

@@ -6,13 +6,15 @@ import mlflow
 
 SAMPLE_RATE = 22050
 
-@mlflow.trace
+#@mlflow.trace
 def process_file(file_path):
     """
     Load 10 second chunks single song.
     """
     y, sr = librosa.load(file_path, mono=True, sr=SAMPLE_RATE)
-    size = int(SAMPLE_RATE * 5)
+    if(not sr == SAMPLE_RATE):
+        return []
+    size = int(SAMPLE_RATE * 10)
     sample_len = len(y)
 
     file_chunks = []
@@ -20,18 +22,19 @@ def process_file(file_path):
         end = start_pos + size
         if end <= sample_len:
             chunk = y[start_pos:end]
-            chunk = librosa.feature.melspectrogram(y=chunk,sr=SAMPLE_RATE)
-            chunk = ((librosa.amplitude_to_db(chunk,ref=np.max)+40)/40)
+            #chunk = librosa.feature.melspectrogram(y=chunk,sr=SAMPLE_RATE)
+            #chunk = ((librosa.amplitude_to_db(chunk,ref=np.max)+40)/40)
             file_chunks.append(chunk)
     return file_chunks
 
-@mlflow.trace
+#@mlflow.trace
 def load():
     """
     Load 10 second chunks of songs.
     """
     audio = []
     files = list(Path("./data/").glob("*.mp3"))
+    #files = files[:12]
     with Pool(cpu_count()) as pool:
         chunk_list = pool.map(process_file, files)
     for l in chunk_list:
@@ -52,15 +55,12 @@ def audio_split(audio):
     x = librosa.feature.melspectrogram(y=x, sr=SAMPLE_RATE)
     y = librosa.feature.melspectrogram(y=y, sr=SAMPLE_RATE)
 
-    ma,mi = x.max(), x.min()
-    x = (x - mi) / (ma - mi)
-            
-    ma,mi = y.max(), y.min()
-    y = (y - mi) / (ma - mi)
+    x = ((librosa.amplitude_to_db(x,ref=np.max)+80)/80)
+    y = ((librosa.amplitude_to_db(y,ref=np.max)+80)/80)
     
     return x,y
 
-def detaset(chunks):
+def dataset(chunks):
     """
     convert 10 second chunks to dataset
     """