icefall
icefall copied to clipboard
k2-fsa
I only tested it on Switchboard alone, but ran the data prep for Fisher and I expect it to work OK. We don't have eval2000 data prep yet so it's...
This recipe implements Hubert transducer model. It supports finetuning a pretrained Hubert model with custom vocabulary using pruned rnnt loss. To use this recipe, you need fairseq as dependency. The...
I used my own data to train a streaming model. The recognition effect is poor when decoding. There are two obvious problems, one is to delete words at the end,...
# WER comparison | | test-clean| test-other| comment| |---|---|---|---| |basline| 2.78 | 7.36 | --iter468000 --avg 16, greedy search| |this PR|3.36|8.20| --iter 468000 --avg 16, greedy search| It performs better...
In the commit (https://github.com/k2-fsa/icefall/pull/166/commits/b49510e2bf7064f4f60650e6787288db1bad2941), icefall has been implemented label smoothing for transducer models, but now it has been removed. Why does icefall stop supporting label smoothing in transducer models.
For most speech datasets, we have already extracted their fbank features by `compute-fbank-feats` of Kaldi. Is it possible to generate the `(dataset_name)_cuts_train.jsonl.gz` directly using Kaldi's various List ( wav.scp, utt2spk,...
Hi authors, Very amazing work! Is there any scripts to prepare a new dataset that I can train or test? Or to simplify, how can I test the wer of...
Hi, there. I was trying to resume the training from a checkpoint `checkpoint-28000.pt`. But it looks like the train sampler iterates until `StopIteration` ``` 2022-08-24 23:20:54,372 INFO [train.py:940] (3/4) Training...
