Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Dec 20, 2019

• update to pytorch1.2 and python3

  Diamondfan

  committed
  eddd255

  Copy full SHA for eddd255

Commits on May 1, 2018

• use ark directly to simplify the data_loader.py and add waveform as input

  

  Richardfan

  committed
  90403b9

  Copy full SHA for 90403b9

Commits on Apr 2, 2018

• add model file

  

  Richardfan

  committed
  eb17189

  Copy full SHA for eb17189

Commits on Mar 30, 2018

• adapt to the new structure of data, can run with no error

  

  Richardfan

  committed
  70a1f60

  Copy full SHA for 70a1f60

Commits on Mar 29, 2018

• Add data prepare parts and extract feats using kaldi!

  

  Richardfan

  committed
  a6fc555

  Copy full SHA for a6fc555

Commits on Mar 20, 2018

• simplify some process

  

  Richardfan

  committed
  afceaec

  Copy full SHA for afceaec

Commits on Jan 4, 2018

• re-organize the code

  

  Richardfan

  committed
  36812e5

  Copy full SHA for 36812e5

• re-organize code, combine two models into one and can adjust the cnn param in the conf file

  

  Richardfan

  committed
  5fc602e

  Copy full SHA for 5fc602e

Commits on Dec 25, 2017

• change observation to dev loss

  

  Richardfan

  committed
  024c2a6

  Copy full SHA for 024c2a6

Commits on Dec 21, 2017

• add phoneme-level bi-gram LM for beam search decoding

  

  Richardfan

  committed
  d3e3f74

  Copy full SHA for d3e3f74

Commits on Dec 5, 2017

• README.md

  

  Richardfan

  committed
  8d8d081

  Copy full SHA for 8d8d081

• README.md

  

  Richardfan

  committed
  af02c8e

  Copy full SHA for af02c8e

• drop the ctcdecode part for errors, use python implemention of beamsearch for decode

  

  Richardfan

  committed
  d176305

  Copy full SHA for d176305

Commits on Nov 30, 2017

• add visualize for cnn_lstm_ctc

  

  Richardfan

  committed
  52a46dc

  Copy full SHA for 52a46dc

• change a better way to adjust learning rate

  

  Richardfan

  committed
  72edc94

  Copy full SHA for 72edc94

Commits on Nov 23, 2017

• add random seed for gpu to make experiment reproducible

  

  Richardfan

  committed
  dba622c

  Copy full SHA for dba622c

Commits on Nov 16, 2017

• add utils.py to simplify the code

  

  Richardfan

  committed
  93368a6

  Copy full SHA for 93368a6

Commits on Nov 10, 2017

• make the code same with local

  

  Richardfan

  committed
  28ef056

  Copy full SHA for 28ef056

Commits on Nov 8, 2017

• add deep copy for bug of saving best model

  

  Richardfan

  committed
  415ac1e

  Copy full SHA for 415ac1e

Commits on Nov 6, 2017

• change the training process for 863 corpus

  

  Richardfan

  committed
  a6b2e94

  Copy full SHA for a6b2e94

Commits on Nov 2, 2017

• map 48 phones to 39 when decoding

  

  Richardfan

  committed
  6bed5be

  Copy full SHA for 6bed5be

Commits on Nov 1, 2017

• add my_863_corpus

  

  Richardfan

  committed
  a6402ba

  Copy full SHA for a6402ba

• change structure of dir to add different dataset

  

  Richardfan

  committed
  a3c3882

  Copy full SHA for a3c3882

Commits on Oct 26, 2017

• fix bugs for BeamSearchDecoder, now it works

  

  Richardfan

  committed
  fbe9152

  Copy full SHA for fbe9152

Commits on Oct 18, 2017

• add CNN before RNN with Spectrum input

  

  Richardfan

  committed
  79f8acd

  Copy full SHA for 79f8acd

Commits on Oct 16, 2017

• deal input for CNN

  

  Richardfan

  committed
  6cd1780

  Copy full SHA for 6cd1780

Commits on Sep 18, 2017

• read the model parameters from saved pkl file

  

  Richardfan

  committed
  8db7ad5

  Copy full SHA for 8db7ad5

• fix the order of dataloader

  

  Richardfan

  committed
  b682182

  Copy full SHA for b682182

• add spectrum extraction for data prepare

  

  Richardfan

  committed
  aec08dd

  Copy full SHA for aec08dd

Commits on Sep 16, 2017

• fix beam decoder, but still no improve

  

  Richardfan

  committed
  5b64fd1

  Copy full SHA for 5b64fd1

Commits on Sep 15, 2017

• add phonemes recognition on the base of character

  

  Richardfan

  committed
  9ed7d31

  Copy full SHA for 9ed7d31

Commits on Sep 14, 2017

• extract feats and apply global cmvn from kaldi toolkit

  

  Richardfan

  committed
  08cf451

  Copy full SHA for 08cf451

• fix the way of adjusting learning rate

  

  Richardfan

  committed
  af2deac

  Copy full SHA for af2deac

• update README

  

  Richardfan

  committed
  c186eb2

  Copy full SHA for c186eb2

• add Beamdecoder

  

  Richardfan

  committed
  88c11e5

  Copy full SHA for 88c11e5

Pagination

Previous

Next