switched to gan and train.py.

data.py

@@ -2,15 +2,17 @@ import librosa
 import numpy as np
 from pathlib import Path
 from multiprocessing import Pool, cpu_count
+import mlflow
 
 SAMPLE_RATE = 22050
 
+@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 * 10)
+    size = int(SAMPLE_RATE * 5)
     sample_len = len(y)
 
     file_chunks = []
@@ -18,9 +20,12 @@ 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)
             file_chunks.append(chunk)
     return file_chunks
 
+@mlflow.trace
 def load():
     """
     Load 10 second chunks of songs.
@@ -33,6 +38,9 @@ def load():
         audio.extend(l)
     return audio
 
+
+
+##DEP
 def audio_split(audio):
     """
     Split 10 seconds of audio to 2 5 second clips