Vietnamese Automatic Speech Recognition with Transformer

EasyChair Preprint 7147

Abstract

Recently, speech recognition using end-toend models is gradually becoming a trend and has superior performance compared to traditional methods. The most frequently used methods are the combination of attention-based methods use an attention mechanism and connectionist temporal classification (CTC) for supervised Learning for Automatic Speech Recognition (ASR). In this paper, we propose a speech recognition model using the transformer architecture and achieved the top 3 in 2021 the Vietnamese Language and Speech Processing contest with 8.83% word error rate (WER) on private-test set.

Keyphrases: Attention Mechanism, end-to-end speech recognition, transformer

@booklet{EasyChair:7147,
  author    = {Duong Trinh Anh and Sam Dang Van and Tuan Do Van and Vi Ngo Van},
  title     = {Vietnamese Automatic Speech Recognition with Transformer},
  howpublished = {EasyChair Preprint 7147},
  year      = {EasyChair, 2021}}