* Add utility for shallow fusion
* test batch size == 1 without shallow fusion
* Use shallow fusion for modified-beam-search
* Modified beam search with ngram rescoring
* Fix code according to review
Co-authored-by: Fangjun Kuang <csukuangfj@gmail.com>
* Sort result to make it more convenient to compare decoding results
* Add cut_id to recognition results
* add cut_id to results for all recipes
* Fix torch.jit.script
* Fix comments
* Minor fixes
* Fix torch.jit.tracing for Pytorch version before v1.9.0
* add stats about duration and padding proportion
* add for utt_duration
* add stats for other recipes
* add stats for other 2 recipes
* modify doc
* minor change
* add pruned transducer stateless5 recipe for tal_csasr
* do some changes for merging
* change for conformer.py
* add wer and cer for Chinese and English respectively
* fix a error for conformer.py
* support streaming in conformer
* Add more documents
* support streaming on pruned_transducer_stateless2; add delay penalty; fixes for decode states
* Minor fixes
* streaming for pruned_transducer_stateless4
* Fix conv cache error, support async streaming decoding
* Fix style
* Fix style
* Fix style
* Add torch.jit.export
* mask the initial cache
* Cutting off invalid frames of encoder_embed output
* fix relative positional encoding in streaming decoding for compution saving
* Minor fixes
* Minor fixes
* Minor fixes
* Minor fixes
* Minor fixes
* Fix jit export for torch 1.6
* Minor fixes for streaming decoding
* Minor fixes on decode stream
* move model parameters to train.py
* make states in forward streaming optional
* update pretrain to support streaming model
* update results.md
* update tensorboard and pre-models
* fix typo
* Fix tests
* remove unused arguments
* add streaming decoding ci
* Minor fix
* Minor fix
* disable right context by default
* Add fast_beam_search_nbest.
* Fix CI errors.
* Fix CI errors.
* More fixes.
* Small fixes.
* Support using log_add in LG decoding with fast_beam_search.
* Support LG decoding in pruned_transducer_stateless
* Support LG for pruned_transducer_stateless2.
* Support LG for fast beam search.
* Minor fixes.
* Use jsonl for cutsets in the librispeech recipe.
* Use lazy cutset for all recipes.
* More fixes to use lazy CutSet.
* Remove force=True from logging to support Python < 3.8
* Minor fixes.
* Fix style issues.
* keep model_avg on cpu
* explicitly convert model_avg to cpu
* minor fix
* remove device convertion for model_avg
* modify usage of the model device in train.py
* change model.device to next(model.parameters()).device for decoding
* assert params.start_epoch>0
* assert params.start_epoch>0, params.start_epoch
* First upload of model average codes.
* minor fix
* update decode file
* update .flake8
* rename pruned_transducer_stateless3 to pruned_transducer_stateless4
* change epoch number counter starting from 1 instead of 0
* minor fix of pruned_transducer_stateless4/train.py
* refactor the checkpoint.py
* minor fix, update docs, and modify the epoch number to count from 1 in the pruned_transducer_stateless4/decode.py
* update author info
* add docs of the scaling in function average_checkpoints_with_averaged_model
* initial commit
* support download, data prep, and fbank
* on-the-fly feature extraction by default
* support BPE based lang
* support HLG for BPE
* small fix
* small fix
* chunked feature extraction by default
* Compute features for GigaSpeech by splitting the manifest.
* Fixes after review.
* Split manifests into 2000 pieces.
* set audio duration mismatch tolerance to 0.01
* small fix
* add conformer training recipe
* Add conformer.py without pre-commit checking
* lazy loading and use SingleCutSampler
* DynamicBucketingSampler
* use KaldifeatFbank to compute fbank for musan
* use pretrained language model and lexicon
* use 3gram to decode, 4gram to rescore
* Add decode.py
* Update .flake8
* Delete compute_fbank_gigaspeech.py
* Use BucketingSampler for valid and test dataloader
* Update params in train.py
* Use bpe_500
* update params in decode.py
* Decrease num_paths while CUDA OOM
* Added README
* Update RESULTS
* black
* Decrease num_paths while CUDA OOM
* Decode with post-processing
* Update results
* Remove lazy_load option
* Use default `storage_type`
* Keep the original tolerance
* Use split-lazy
* black
* Update pretrained model
Co-authored-by: Fangjun Kuang <csukuangfj@gmail.com>
