search

hirofumi0810 / neural_sp

End-to-end ASR/LM implementation with PyTorch
Apache License 2.0
589 stars 140 forks source link

Add batch size type for efficient training #312

Closed hirofumi0810 closed 3 years ago

hirofumi0810 commented 3 years ago

Batch size is dynamically determined based on the total number of input frames.

codecov-io commented 3 years ago

Codecov Report

Merging #312 (fe3b593) into master (bf1d5a4) will increase coverage by 0.03%. The diff coverage is 85.71%.

Impacted file tree graph

@@            Coverage Diff             @@
##           master     #312      +/-   ##
==========================================
+ Coverage   67.78%   67.81%   +0.03%     
==========================================
  Files          96       96              
  Lines       10835    10863      +28     
==========================================
+ Hits         7344     7367      +23     
- Misses       3491     3496       +5
Impacted Files Coverage Δ
neural_sp/bin/asr/train.py 65.31% <ø> (ø)
neural_sp/evaluators/phone.py 10.11% <0.00%> (ø)
neural_sp/evaluators/wordpiece_bleu.py 10.00% <0.00%> (ø)
neural_sp/datasets/utils.py 62.03% <84.21%> (+3.70%) :arrow_up:
neural_sp/bin/args_asr.py 96.56% <100.00%> (+0.01%) :arrow_up:
neural_sp/datasets/asr/build.py 55.17% <100.00%> (ø)
neural_sp/datasets/asr/dataloader.py 100.00% <100.00%> (ø)
neural_sp/datasets/asr/sampler.py 86.11% <100.00%> (+0.60%) :arrow_up:
neural_sp/evaluators/character.py 92.19% <100.00%> (ø)
neural_sp/evaluators/word.py 56.25% <100.00%> (ø)
... and 3 more

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update bf1d5a4...fe3b593. Read the comment docs.