icefall

icefall copied to clipboard

Combines #560 and #558

csukuangfj

Clip rnn gradients in a chunk-wise manner

1

This PR aims to clip the rnn gradients in a chunk-wise manner, to solve the gradient explosion problem in the backward pass. When computing each chunk, we clip the gradients...

yaozengwei

I'm planning to implement some SOTA speech synthesis models on LJSpeech and LibriTTS, any suggestions?

4

I'm new in k2, but very familiar with Kaldi(I wrote the [kaldi-ctc](https://github.com/lingochamp/kaldi-ctc)). Text Analyzer - In the first stage, it will be simple. - In the future, it will include...

lifeiteng

Librispeech assert x.size(0) == lengths.max().item()

4

I successfully trained a librispeech-100 pruned_transducer_stateless2 model. But when I tried to train on my own data I got an exception. 2022-08-30 13:45:32,285 INFO [train.py:1026] features shape: torch.Size([3, 1939, 80])...

dfmordol

This PR uses CTC as an auxiliary loss for the streaming LSTM (see https://github.com/k2-fsa/icefall/pull/479) transducer model, following https://github.com/k2-fsa/icefall/pull/477.

yaozengwei

GigaSpeech Streaming Model

1

wgb14

RuntimeError: received 0 items of ancdata

1

When using prepare.sh file in wenetspeech from stage 1, a runtime error raised during computer features of `data/fbank/cuts_TEST_NET_raw.jsonl.gz`. Here is the logs: ``` 2022-08-02 22:39:27,066 INFO [compute_fbank_wenetspeech_dev_test.py:63] Loading data/fbank/cuts_TEST_NET_raw.jsonl.gz 2022-08-02...

JaheimLee

WIP: Transducer distill

2

2 step distill, through CE loss directly by CTC/Transducer label and Transducer output tensor without extra CTC and Transducer loss: 1. CTC model-> Transducer model 2. Transducer model 1 ->...

qmpzzpmq

Very high WER when I tried to use bpe_500 lexicon on tdnn_lstm_ctc model

2

Yes, I know tdnn-lstm is kinda outdated but it's easier to fit it to our edge device, so I thought it might be worth trying. When I trained it the...

wangtiance

Performance degradation when training with jsonl manifests (i.e. DynamicBucketingSampler)

4

Create a issue here to discuss this problem. I re-ran the pruned_transducer_stateless4 baseline with `DynamicBucketingSampler`, the results are as follows: setup | greedy_search | fast_beam_search | modified_beam_search | comments --...

pkufool