snowfall
snowfall copied to clipboard
k2-fsa
Moved to https://github.com/k2-fsa/icefall
I thought it makes sense to archive this repo in GitHub as it's not being actively developed anymore. It will make it read-only and clearly display it's archived to everybody...
Hi experts, I was wondering if the build_ctc_topo interface can be used for blankid is not equal to 0? What's the correct way to build the ctc_topo in this situation?...
I set subsampling_factor=2. #TIMIT CTC_Train Testing results based on different training epochs: ``` epoch=20 2021-09-03 10:54:10,903 INFO [ctc_decode.py:188] %PER 30.34% [2225 / 7333, 293 ins, 441 del, 1491 sub ]...
Hi, team! I have encountered a problem with k2 in my code. Below is the description of this problem. For a `nnet_output` with shape `[B, T, D]`, I am trying...
1. A better model trained by (ctc + label_smooth_loss #219) is released 2. 4-gram rescore is integrated with refering to #215 Latest result with feat_batch_norm | Wer% on test_clean |...
See the code below: https://github.com/k2-fsa/snowfall/blob/350253144af04c295f560cdb976f817dc13b2993/snowfall/models/transformer.py#L162 https://github.com/k2-fsa/snowfall/blob/350253144af04c295f560cdb976f817dc13b2993/snowfall/models/transformer.py#L167 https://github.com/k2-fsa/snowfall/blob/350253144af04c295f560cdb976f817dc13b2993/snowfall/models/transformer.py#L179 https://github.com/k2-fsa/snowfall/blob/350253144af04c295f560cdb976f817dc13b2993/snowfall/models/transformer.py#L709 https://github.com/k2-fsa/snowfall/blob/350253144af04c295f560cdb976f817dc13b2993/snowfall/models/transformer.py#L720-L721 ---- You can see that `ys_in_pad` is padded with eos_id, which is a positive word piece ID. However, it is using...
It happens only when `--concatenate-cuts=True`. See the problematic code below (line 692): https://github.com/k2-fsa/snowfall/blob/350253144af04c295f560cdb976f817dc13b2993/snowfall/models/transformer.py#L687-L692 When `--concatenate-cuts=True`, several utterances may be concatenated into one sequence. So `lengths[sequence_idx]` may correspond to multiple utterances....
As metioned in #217，currently bpe training with ctcLoss and labelSmoothLoss in snowfall obtain higher wer than that of espnet. decoding algorithm| training tool| encoder + k2 ctc decode+no rescore |...
Just throwing the idea around in case somebody wants to pick it up, I wanted to do it for some time but can't find the time. There's a new position...
[Guys, I have gym now so I'll submit this and write the rest of this later today. ] I am creating an issue to describe a plan for multi-pass n-best-list...
