Ma

Hi, Think to your paper and this github, I have re-implemented this project. but my final result is 3.5 of clean and 9.0 of other in WER, the final loss...

dtype mismatch in AttentiveStatisticsPooling with FP16 training mode

1

### Describe the bug https://github.com/speechbrain/speechbrain/blob/5beaece0b4cdcce18f303cbb1f5f04b0a879dca2/speechbrain/lobes/models/ECAPA_TDNN.py#L275 If the model is trained with FP16 or BF16 mode, here will report dtype mismatch. So, one solution is that it need add `.to(x.dtype)`. ###...

Batch with multi audio input.

2

Hi, I see the batched inference is for the long audio. If the input is multi-short audio, could it support the batched inference? Thank you in advance.