convert audio samples to int16

app.py

@@ -31,6 +31,10 @@ speaker_embeddings = {
 }
 
 
+target_dtype = np.int16
+max_range = np.iinfo(target_dtype).max
+
+
 def predict(text, speaker, post_process):
     if len(text.strip()) == 0:
         return (16000, np.zeros(0))
@@ -49,9 +53,13 @@ def predict(text, speaker, post_process):
     speech = pipe(text, forward_params=forward_params)
 
     if post_process:
-        return enhance_audio(speech["audio"], speech["sampling_rate"], device)
+        sr, audio = enhance_audio(speech["audio"], speech["sampling_rate"], device)
     else:
-        return (speech["sampling_rate"], speech["audio"])
+        sr, audio = speech["sampling_rate"], speech["audio"]
+
+    audio = (audio * max_range).astype(np.int16)
+
+    return sr, audio
 
 
 def replace_danish_letters(text):