An Effective Training Framework for Light-Weight Automatic Speech Recognition Models

Official repo for the paper "An Effective Training Framework for Light-Weight Automatic Speech Recognition Models" accepted at InterSpeech 2025.

The paper is available etf4asr

Update

Code is released !!!

Find the Baseline model for LibriSpeech dataset that is used as a reference model in our experimentation. model

Name		Name	Last commit message	Last commit date
Latest commit History 7 Commits
models		models
util		util
1		1
2		2
NoamOpt.py		NoamOpt.py
README.md		README.md
conf.py		conf.py
dataloaders.py		dataloaders.py
evaluate.py		evaluate.py
inference.py		inference.py
levenshtein		levenshtein
lexicon.txt		lexicon.txt
libri.bpe-256.model		libri.bpe-256.model
libri.bpe-256.vocab		libri.bpe-256.vocab
librispeech-bpe-256.lex		librispeech-bpe-256.lex
librispeech-bpe-256.tok		librispeech-bpe-256.tok
librispeech.lex		librispeech.lex
pretrain_for_ASR.py		pretrain_for_ASR.py
pruning.py		pruning.py
pruning_inference.py		pruning_inference.py
tokens.txt		tokens.txt
train.py		train.py
transforms.py		transforms.py
wer.sh		wer.sh