Code

[WIP] Lhotse/K2 example#45

Code

Open

freewym wants to merge 119 commits intomainfreewym/espresso:mainfrom lhotsefreewym/espresso:lhotseCopy head branch name to clipboard

+15,419-189

Commits

Commits on Dec 24, 2020

• initial commit

  freewym

  committed
  5d1001d

  Copy full SHA for 5d1001d

• asr models related

  freewym

  committed
  dbb20ad

  Copy full SHA for dbb20ad

• decoding related

  freewym

  committed
  d6033af

  Copy full SHA for d6033af

• code adaptation/changes according to the commits from Dec 24, 2018 to Jan 9, 2019

  freewym

  committed
  eadf2a8

  Copy full SHA for eadf2a8

• wsj recipe and other fixes

  freewym

  committed
  6148a8a

  Copy full SHA for 6148a8a

• code adaptation/changes according to the commits from Jan 24, 2019 to Jan 25, 2019

  freewym

  committed
  ee70584

  Copy full SHA for ee70584

• fix

  freewym

  committed
  93c1de4

  Copy full SHA for 93c1de4

• validation on wer

  freewym

  committed
  e864565

  Copy full SHA for e864565

• environment configurations

  freewym

  committed
  8782e5d

  Copy full SHA for 8782e5d

• code adaptation/changes according to the commits from Jan 29, 2019 to Feb 9, 2019

  freewym

  committed
  9363a71

  Copy full SHA for 9363a71

• Add wsj data prep recipe from kaldi and espnet

  freewym

  committed
  0802783

  Copy full SHA for 0802783

• code adaptation/changes according to the commits from Feb 12, 2019 to Feb 22, 2019

  freewym

  committed
  d276c82

  Copy full SHA for d276c82

• code adaptation/changes according to the commits from Feb 23, 2019 to Mar 4, 2019

  freewym

  committed
  c1a54da

  Copy full SHA for c1a54da

• add lr scheduler that allows to set epoch from which lr starts to decay

  freewym

  committed
  47dc107

  Copy full SHA for 47dc107

• refactor speech_tools.utils.Tokenizer

  freewym

  committed
  eb1d55f

  Copy full SHA for eb1d55f

• lm training

  freewym

  committed
  f0930b1

  Copy full SHA for f0930b1

• code adaptation/changes according to the commits from Apr 1, 2019 to Apr 10, 2019

  freewym

  committed
  0e2926b

  Copy full SHA for 0e2926b

• word lm related; add unigram/temporal label smoothing and update the wsj

  Show description for 9e479ca
  freewym

  committed
  9e479ca

  Copy full SHA for 9e479ca

• add scheduled sampling training support

  freewym

  committed
  cd8803d

  Copy full SHA for cd8803d

• add bpe support & librispeech recipe

  freewym

  committed
  82a1175

  Copy full SHA for 82a1175

• pure batch decoding with LookAhead Word LM

  freewym

  committed
  3ea6bb9

  Copy full SHA for 3ea6bb9

• librispeech recipe fix; code adaptation/changes according to the commits on June 11, 2019

  freewym

  committed
  36baed5

  Copy full SHA for 36baed5

• add Transformer and FConv model for ASR

  freewym

  committed
  f4bcc7f

  Copy full SHA for f4bcc7f

• LM arch changes

  freewym

  committed
  9eb1020

  Copy full SHA for 9eb1020

• swbd

  

  Hang Lyu

  authored and

  freewym

  committed
  9745aa7

  Copy full SHA for 9745aa7

• add coverage to libripspeech recipe; make best metric for ASR configurable by using --best-checkpoint-metric; code adaptation/changes according to the commits from Jun 26, 2019 to Jul 3, 2019

  freewym

  committed
  7ae0222

  Copy full SHA for 7ae0222

• fix swbd recipe; code adaptation/changes according to the commits from Jul 17, 2019 to Jul 19, 2019

  freewym

  committed
  6ab3bd5

  Copy full SHA for 6ab3bd5

• revise speech_{transformer/fconv} code

  freewym

  committed
  236c927

  Copy full SHA for 236c927

• improve swbd recipe; comtinue training while lr is no less than --min-lr; code adaptation/changes according to the commits from Jul 24, 2019 to Aug 1, 2019

  freewym

  committed
  3b38acd

  Copy full SHA for 3b38acd

• Relicense Espresso under MIT license

  freewym

  committed
  833ad03

  Copy full SHA for 833ad03

• add --eos-factor for beam search to alleviate the problem of too short transcripts with LM fusion

  freewym

  committed
  2777ac7

  Copy full SHA for 2777ac7

• tokenize each sentence such that it ends with <space>; modify look-ahead LM accordingly; modify the WSJ recipe accordingly

  freewym

  committed
  5cf1c41

  Copy full SHA for 5cf1c41

• switch to pip install sentencepiece; modify Librispeech/SWBD recipes accordingly; code adaptation/changes according to the commits on Aug 21 and 30, 2019

  freewym

  committed
  b2d3ef2

  Copy full SHA for b2d3ef2

• update tensorized tree implementation

  

  ctongfei

  authored and

  freewym

  committed
  1c19fe5

  Copy full SHA for 1c19fe5

• switch to pip install kaldi_io

  freewym

  committed
  f21c428

  Copy full SHA for f21c428

• Update README.md; add logo; slightly change LM weight and beam size for

  Show description for bac7415
  

  ctongfei

  authored and

  freewym

  committed
  bac7415

  Copy full SHA for bac7415

• compansate for the removal of torch.rand() from distributed_init() recently introduced in fairseq, to make ASR results reproducible

  freewym

  committed
  91d9fc7

  Copy full SHA for 91d9fc7

• add backoff smoothing for unigram label smoothing

  freewym

  committed
  6a8f4d3

  Copy full SHA for 6a8f4d3

• a better LM for Librispeech yielding better WERs; code

  Show description for 32825b5
  freewym

  committed
  32825b5

  Copy full SHA for 32825b5

• code adaptation/changes according to the commits on Sep 30, 2019

  freewym

  committed
  c63caa8

  Copy full SHA for c63caa8

• set --distributed-port=-1 if ngpus=1; code adaptation/changes according to the commits on Oct 11, 2019

  freewym

  committed
  0e610d5

  Copy full SHA for 0e610d5

• change warmup scheduling for ReduceLROnPlateauV2; code adaptation/changes according to the commits on Oct 18, 2019

  freewym

  committed
  67fdff1

  Copy full SHA for 67fdff1

• remove warmup code in ReduceLROnPlateauV2 as fariseq just added it; code adaptation/changes according to the commits on Oct 23, 2019

  freewym

  committed
  aab85e5

  Copy full SHA for aab85e5

• add gpu.conf for SGE qsub

  freewym

  committed
  3018d20

  Copy full SHA for 3018d20

• Fixed error when using fp16

  Show description for a1b76df
  

  Shujian2015

  authored and

  freewym

  committed
  a1b76df

  Copy full SHA for a1b76df

• code adaptation/changes according to the commits on Nov 8, 2019

  freewym

  committed
  75581ae

  Copy full SHA for 75581ae

• code adaptation/changes according to the commits on Nov 26, 2019

  freewym

  committed
  0e91c3b

  Copy full SHA for 0e91c3b

• code adaptation/changes according to the commits on Dec 11, 2019

  freewym

  committed
  7f3a0bc

  Copy full SHA for 7f3a0bc

• allows text2vocabulary.py to accept an existing vocabulry with its first column as words

  freewym

  committed
  5744fb9

  Copy full SHA for 5744fb9

• re-organize the codebase to isolate espresso from fairseq

  freewym

  committed
  4617036

  Copy full SHA for 4617036

• remove coverage term for beam search decoding as it has been superceded by eos thresholding

  freewym

  committed
  9dd8992

  Copy full SHA for 9dd8992

• fix bugs causing build failure; a bunch of lint changes; rename

  Show description for 69dbd15
  freewym

  committed
  69dbd15

  Copy full SHA for 69dbd15

• scheduled sampling rate scheduler

  

  ctongfei

  authored and

  freewym

  committed
  732ab08

  Copy full SHA for 732ab08

• decouple scheduled sampling rate scheduler; rename all appearances of "dict" variables to "dictionary" as "dict" is a reserved keyword in Python; affect swbd results due to some numerical issue of …

  Show description for ad0157e
  freewym

  committed
  ad0157e

  Copy full SHA for ad0157e

• code adaptation/changes according to the commits from Dec 21 to Dec 24, 2019

  freewym

  committed
  13e2d62

  Copy full SHA for 13e2d62

• move the code of computing prob mask of temporal label smoothing into a separate function

  freewym

  committed
  390d157

  Copy full SHA for 390d157

• isolate greedy search code from criterions (#19)

  Show description for 3c5b3c9
  freewym

  committed
  3c5b3c9

  Copy full SHA for 3c5b3c9

• remove the need to pass tgt dataset to criterions by adding a raw text field to collated samples

  Show description for 54870cb
  freewym

  committed
  54870cb

  Copy full SHA for 54870cb

• code adaptation/changes according to the commits from Jan 3 to Jan 6, 2020

  freewym

  committed
  3677002

  Copy full SHA for 3677002

• code adaptation/changes according to the commits from Jan 11 to Jan 14, 2020

  freewym

  committed
  f1bed6f

  Copy full SHA for f1bed6f

• code adaptation/changes according to the commits from Jan 16 to Jan 17, 2020; move decode log files to decode result dirs

  freewym

  committed
  458828d

  Copy full SHA for 458828d

• move WER validation with greedy decoding code from criterion to task; valid loss is now based on teacher enforcing instead of greedy decoding; rename {,label_smoothed_}cross_entropy_with_wer.py to …

  Show description for 293a068
  freewym

  committed
  293a068

  Copy full SHA for 293a068

• code adaptation/changes according to the commits on Jan 20, 2020; cosmetic changes for lookahead LM

  freewym

  committed
  80f564d

  Copy full SHA for 80f564d

• isolate LSTMLanguageModel from speech_lstm.py and rename it to LSTMLanguageModelEspresso to avoid naming conflicts with the fairseq\'s LSTMLanguageModel introduced recently

  freewym

  committed
  d488b92

  Copy full SHA for d488b92

• code adaptation/changes according to the commits on Jan 30, 2020

  freewym

  committed
  ad102bd

  Copy full SHA for ad102bd

• code adaptation/changes according to the commits on Feb 12, 2020

  freewym

  committed
  19ec973

  Copy full SHA for 19ec973

• add options to accept utt2num_frames files to speed up the data loading (#22)

  Show description for 9c69029
  freewym

  committed
  9c69029

  Copy full SHA for 9c69029

• use json files to simplify the cli options for input data (#23)

  freewym

  committed
  698305f

  Copy full SHA for 698305f

• code adaptation/changes according to the commits on Feb 21, 2020

  freewym

  committed
  ea7732b

  Copy full SHA for ea7732b

• move duplicated network parsers to espresso/speech_tools/utils.py; remove remaining coverage option passed in wsj recipe

  freewym

  committed
  7170edc

  Copy full SHA for 7170edc

• code adaptation/changes according to the commits on Feb 27-29, 2020

  freewym

  committed
  95fcd90

  Copy full SHA for 95fcd90

• code adaptation/changes according to the commits on Mar 3-10, 2020

  freewym

  committed
  41c5725

  Copy full SHA for 41c5725

• SpecAugment (#21)

  freewym

  committed
  4b8d3be

  Copy full SHA for 4b8d3be

• code adaptation/changes according to the commits on Mar 11, 2020; change

  Show description for 11e106f
  freewym

  committed
  11e106f

  Copy full SHA for 11e106f

• fix specaug indexing

  freewym

  committed
  c05a336

  Copy full SHA for c05a336

• code adaptation/changes according to the commits on Mar 24-Apr 3, 2020; use data.encoders.{bpe,tokenizer} for wordpiece decode

  freewym

  committed
  8772f69

  Copy full SHA for 8772f69

• code adaptation/changes according to the commits on Apr 7

  freewym

  committed
  e40b033

  Copy full SHA for e40b033

• update the qsub script for gpu jobs; code adaptation/changes according to the commits on Apr 16

  freewym

  committed
  b9c19c2

  Copy full SHA for b9c19c2

• use EncoderOut for SpeechLSTMEncoder's output; code adaptation/changes according to the commits on Apr 21

  freewym

  committed
  b35d210

  Copy full SHA for b35d210

• Hybrid ASR code (E2E LF-MMI and cross-entropy) and WSJ examples (#29)

  freewym

  committed
  cfa899c

  Copy full SHA for cfa899c

• code adaptation/changes according to the commits on May 10

  freewym

  committed
  be4fd6a

  Copy full SHA for be4fd6a

• code adaptation/changes according to the commits on May 18

  freewym

  committed
  967dd34

  Copy full SHA for 967dd34

• fix lf-mmi loss; code adaptation/changes according to the commits on May 27

  freewym

  committed
  af5a564

  Copy full SHA for af5a564

• remove useless max_{source,target}_positions arguments

  freewym

  committed
  9d3b8b3

  Copy full SHA for 9d3b8b3

• code adaptation/changes according to the commits on Jun 18-23, 2020

  freewym

  committed
  c8c1dfb

  Copy full SHA for c8c1dfb

• code adaptation/changes according to the commits on Jun 24-25, 2020; fix

  Show description for 027595b
  freewym

  committed
  027595b

  Copy full SHA for 027595b

• Update Transformer models (#31)

  Show description for e42ec43
  freewym

  committed
  e42ec43

  Copy full SHA for e42ec43

• ignore flake8's FileNotFoundError for soft links to kaldi files; code adaptation/changes according to the commits on Jul 8, 2020

  freewym

  committed
  8600f76

  Copy full SHA for 8600f76

• code adaptation/changes according to the commits on Jul 14, 2020

  freewym

  committed
  bc6901b

  Copy full SHA for bc6901b

• code adaptation/changes according to the commits on Jul 16, 2020

  freewym

  committed
  7c21d3d

  Copy full SHA for 7c21d3d

• code adaptation/changes according to the commits on Jul 20-25, 2020

  freewym

  committed
  4b253b1

  Copy full SHA for 4b253b1

• fix reorder_encoder_out in SpeechChunkTransformerEncoder; code adaptation/changes according to the commits on Jul 28, 2020

  freewym

  committed
  d2f264b

  Copy full SHA for d2f264b

• reorder the elements of the returned tuple of TdnnModel.forward();

  Show description for a92409c
  freewym

  committed
  a92409c

  Copy full SHA for a92409c

• updates for new PyChain (#37)

  Show description for c73c4de
  freewym

  committed
  c73c4de

  Copy full SHA for c73c4de

• code adaptation/changes according to the commits on Aug 10-18, 2020

  freewym

  committed
  afaef92

  Copy full SHA for afaef92

• code adaptation/changes according to the commits on Aug 20-24, 2020

  freewym

  committed
  5351152

  Copy full SHA for 5351152

• code adaptation/changes according to the commits on Aug 31, 2020

  freewym

  committed
  9d71078

  Copy full SHA for 9d71078

• code adaptation/changes according to the commits on Sep 9-11, 2020

  freewym

  committed
  e34b27d

  Copy full SHA for e34b27d

• code adaptation/changes according to the commits on Sep 17-26, 2020

  freewym

  committed
  3fe02ae

  Copy full SHA for 3fe02ae

• code adaptation/changes according to the commits on Oct 1, 2020

  freewym

  committed
  8c02b45

  Copy full SHA for 8c02b45

• code adaptation/changes according to the commits on Oct 2-15, 2020

  freewym

  committed
  707ce3a

  Copy full SHA for 707ce3a

• code adaptation/changes according to the commits on Oct 18-Nov 3, 2020 (lots of changes, mostly for adapting to hydra configs and code formatting)

  freewym

  committed
  262c0a2

  Copy full SHA for 262c0a2

• code adaptation/changes according to the commits on Nov 4-9, 2020

  freewym

  committed
  513b171

  Copy full SHA for 513b171

• code adaptation/changes according to the commits on Nov 11, 2020; obtain

  Show description for 6c6ee41
  freewym

  committed
  6c6ee41

  Copy full SHA for 6c6ee41

• fix an error when more than one external LMs are used for shallow fusion

  freewym

  committed
  2b68caf

  Copy full SHA for 2b68caf

• code adaptation/changes according to the commits on Nov 16-20, 2020; fix a

  Show description for 72fa596
  freewym

  committed
  72fa596

  Copy full SHA for 72fa596

• fix length tensor device issue in lf_mmi loss; code adaptation/changes according to the commits on Dec 3-12, 2020

  freewym

  committed
  6b4e571

  Copy full SHA for 6b4e571

• code adaptation/changes according to the commits on Dec 22, 2020

  freewym

  committed
  1f812eb

  Copy full SHA for 1f812eb

Commits on Dec 26, 2020

• Lhotse/K2 support

  freewym

  committed
  b7b8937

  Copy full SHA for b7b8937

• add a data prep example for lhotse

  freewym

  committed
  8926fa3

  Copy full SHA for 8926fa3

• add random split of negatives

  freewym

  committed
  083ea69

  Copy full SHA for 083ea69

• misc fixes

  freewym

  committed
  ecf8423

  Copy full SHA for ecf8423

• k2 training related (not yet done)

  freewym

  committed
  3713dc1

  Copy full SHA for 3713dc1

• fixes

  freewym

  committed
  4fb0d87

  Copy full SHA for 4fb0d87

• f

  freewym

  committed
  dc3874a

  Copy full SHA for dc3874a

• decoding related

  freewym

  committed
  63ceaf2

  Copy full SHA for 63ceaf2

• some changes

  freewym

  committed
  66d84af

  Copy full SHA for 66d84af

Commits on Dec 28, 2020

• fix negative loss

  freewym

  committed
  8a345cb

  Copy full SHA for 8a345cb

• refactor code

  freewym

  committed
  1b05966

  Copy full SHA for 1b05966