fix

ultravox_processing.py

@@ -151,7 +151,7 @@ class UltravoxProcessor(transformers.ProcessorMixin):
             def cnn_out_len(in_len, kernel, stride=1, padding=1, dilation=1):
                 return np.floor((in_len + (2*padding) - (dilation * (kernel - 1)) - 1)/stride + 1)
             nb_encoder_frames = [cnn_out_len(cnn_out_len(feat_len, kernel=3), kernel=3, stride=2) for feat_len in data["audio_len"]]
-            data["audio_token_len"] = [int(np.ceil(x/self.stack_factor)) for x in nb_encoder_frames]
+            data["audio_token_len"] = [int(x//self.stack_factor) for x in nb_encoder_frames]
 
         if text is not None:
             assert isinstance(