Thanks to visit codestin.com
Credit goes to github.com

Skip to content

cottonlove/STT

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

STT

Developing a specialized speech recognition model for Korean regional dialects.

Dataset

AI-Hub에서 제공하는 중·노년층 한국어 방언 데이터 (경상도)를 학습데이터셋으로 선택

‘1인 발화-따라 말하기 유형’ 중 약 30시간, 3.2GB를 학습 및 검증 데이터셋으로 약 8시간, 850MB를 테스트 데이터셋으로 추출하여 구성함.

Setup

실험 환경: 4 NVIDIA-GeForce-RTX-3090 Configuration – Deepspeech2 Number of encoders: 3, 5 Optimizer: Adam Epoch: 50, 40 Batch size: 128, 64 Train:Val = 8:2

CER

Deepspeech2 (#encoders: 3)

  • Pronunciation 전사기준: 0.267
  • Dialect 전사기준: 0.276 Deepspeech2 (#encoders: 5)
  • Pronunciation 전사기준: 0.212
  • Dialect 전사기준: 0.225

Train loss, CER

train

Validation loss, CER

val

Reference

베이스라인 코드는 김수환 님께서 개발해 공개하신 kospeech (https://github.com/sooftware/kospeech) 를 기반으로 하였습니다.

About

Developing a specialized speech recognition model for Korean regional dialects.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages