qzfnihao

math.floor()返回浮点数，用int（math.floor()）

> The TDS CTC model is different from streaming convnets: > > https://github.com/facebookresearch/wav2letter/blob/master/recipes/models/sota/2019/am_arch/am_tds_ctc.arch > https://github.com/facebookresearch/wav2letter/blob/master/recipes/models/sota/2019/librispeech/train_am_tds_ctc.cfg > > TDS CTC is 1.6GB, while sconv is only 400MB. Your results look fine...

I tried more tds layers, but get a litter improvence. V -1 NFEAT 1 0 SAUG 80 27 2 100 1.0 2 PD 0 5 3 C2 1 15 10...

I have realized one mwer for ctc, just n-best mwer on cpu backend with beam search not prefix beam search. But when i finetuned the model trained with transformer+ctc, the...

Which version do you use？it looks like you merge steaming transducer to other espnet, because in this git, there is not invoking recog_v2