feat: add audio normalization for int16 and int32 audio input arrays

app.py

@@ -99,6 +99,16 @@ def predict_language(audio):
     sample_rate, audio_array = audio
     audio_len_sec = len(audio_array) / sample_rate
 
+    # Normalization
+    print("[LOG] Step 1: Normalizing audio...")
+    if audio_array.dtype == np.int16:
+        print("was npint16")
+        audio_array = audio_array.astype(np.float32) / 32768.0
+    elif audio_array.dtype == np.int32:
+        print("was npint32")
+        audio_array = audio_array.astype(np.float32) / 2147483648.0
+    print(f"[LOG] Memory after normalization: {get_mem_usage():.2f} MB")
+
     # Resampling
     if sample_rate != 16000:
         print(f"[LOG] Step 2: Resampling {sample_rate}Hz -> 16000Hz...")