More fixes to handling stereo, add note on wave

app.py

@@ -47,6 +47,9 @@ def main():
             2. Select an audio sample that represents the target voice you want to convert to.
             3. Click the "Convert" button and listen to the result!
 
+            If providing your own audio files, please use WAVE PCM.
+            Service works with 16kHz, 16 bit, mono audio.
+
             If you are interested to plug in Voice Conversion
             service into your own application, don't hesitate to get in touch with us at
             [[email protected]](mailto:[email protected])

vc_service_request.py

@@ -35,8 +35,14 @@ def prepare_audio(audio: Tuple[int, np.ndarray]) -> np.ndarray:
 
     if wav.ndim == 2:
         # average channels
-        wav = np.mean(wav, axis=0, keepdims=False)
-        
+        if wav.shape[0] == 2:
+            wav = np.mean(wav, axis=0, keepdims=False)
+        if wav.shape[1] == 2:
+            wav = np.mean(wav, axis=1, keepdims=False)
+
+    if wav.ndim != 1:
+        return None
+ 
     # ensure proper sampling rate
     if sr != 16000:
         wav = (wav / 32768.0).astype(np.float)
@@ -94,6 +100,8 @@ def vc_service_request(
     """
     src = prepare_audio(source_audio)
     tgt = prepare_audio(target_audio)
+    if not src or not tgt:
+        return
     if len(src) >= 60 * 16000 or len(tgt) >= 30 * 16000:
         # input is way too long, dont return anything
         return