음성과 신호처리(정리중)
-관련 책
*[ratsgo's speech book](https://ratsgo.github.io/speechbook/docs/neuralam/deepspeech)
-
관련 강의
*Connectionist Temporal Classification(CTC) 모델 관련 강의: 토크ON세미나 딥러닝 기반 음성인식 기초 4강
*wave2vec 논문 관련 강의 : [wav2vec: Self-Supervised Learning of Discrete Speech Representations[(https://www.youtube.com/watch?v=mPtyfqWHs3s). -
관련 논문
Wiliam Chan et al. Listen, Attend and Spell arXiv: 1508.01211
Dario Amodei et al. Deep Speech2: End-to-End Speech Recognition in English and Mandarin arXiv: 1512.02595
Jung-Woo Ha et al. ClovaCall: Korean Goal-Oriented Dialog Speech Corpus for Automatic Speech Recognition of Contact (Centers, arXiv: 2004.09367
Soohwan Kim, Seyoung Bae, Cheolhwang Won, Open-source toolkit for end-to-end Korean speech recognition,SIMPAC
Park, Kyubyong and Mulc, Thomas, CSS10: A Collection of Single Speaker Speech Datasets for 10 Languages, arXiv:1903.11269
논문 정리한 깃헙 : awesome-speech-recognition-speech-synthesis-papers
-
관련 블로그 *딥 러닝 음성 인식에 필요한 훈련 데이터를 직접 만들어보자
*LJSpeech
*[LibriSpeech](https://www.openslr.org/11/) : https://paperswithcode.com/sota/speech-recognition-on-librispeech-test-clean
*Libri-Light: 60k hour unlabelled speech + (10h, 1h or 10min) labelled speech (same as LibriVox???) https://github.com/facebookresearch/libri-light
*[KsponSpeech ](https://aihub.or.kr/aidata/105/download
*모두의말뭉치[일상대화_음성_말뭉치](https://corpus.korean.go.kr/)
*[한국어 1인 음성 데이터 ]( https://www.kaggle.com/bryanpark/korean-single-speaker-speech-dataset)
- 신청 접수 후 다운로드 가능
- 철자 전사, 전사 기호(웃음 {laughing}등), 비식별화 기호(이름 &name& 등) 사용
*[Librosa](https://librosa.org/doc/latest/index.html): python 패키지
*Touch Audio : 모델링
*[Kaldi](https://kaldi-asr.org/) : C++로 작성
*[Praat](https://www.fon.hum.uva.nl/praat/)
KoSpeech : https://github.com/sooftware/KoSpeech
참고 링크 : https://pororo-tts.github.io/
wav2vec 결과
영어는 WER (clean 1.9/ other 4.3)
한글은 CER (clean 4.9/ other 5.4)
질문 : CER 계산할때도 grapheme 단위에서 CER 을 구하는 것인지?
관심사
inference : greedy_search