Bug in computing encoder padding mask

[csukuangfj]

It happens only when --concatenate-cuts=True.

See the problematic code below (line 692):

snowfall/snowfall/models/transformer.py

Lines 687 to 692 in 3502531

	for idx in range(supervision_segments.size(0)):
	# Note: TorchScript doesn't allow to unpack tensors as tuples
	sequence_idx = supervision_segments[idx, 0].item()
	start_frame = supervision_segments[idx, 1].item()
	num_frames = supervision_segments[idx, 2].item()
	lengths[sequence_idx] = start_frame + num_frames

When --concatenate-cuts=True, several utterances may be concatenated into one sequence.
So lengths[sequence_idx] may correspond to multiple utterances. Later utterances will OVERWRITE
the value of lengths[sequence_idx] set by earlier utterances if the sequence with sequence_id contains
at least two utterances.

[csukuangfj]

I found this bug while writing tests for encoder_padding_mask. Liyong and I disabled --concatenate-cuts during training,
so it is not a problem for us.