Bug of multi-inputs in attack.py

[oicid]

When I input waves of different lengths, the outputs of the shorter ones have harsh noise. I checked their spectrograms and found the part where the absolute value of the amplitude is less than 0.2 is almost filled. I think it may be caused by the maxlen = max(map(len, audios)) during the processing of the inputs. The shorter ones are then loaded in longer numpy arrays. So I padded the input data so that they have the same length, and then the output is normal. I noticed there is a hint in attack.py (below), and am wondering if it is the bug the author mentioned. Maybe the author can fix it in the next version :)

# Load the inputs that we're given
# TODO: [FINDBUG] loading multiple inputs is possible,
#       but there are some weird things going on at the end of every transcription