악센트DB

  • 설명 :

AccentDB는 구조화되고 레이블이 지정된 악센트 음성의 다중 쌍 병렬 코퍼스입니다. 여기에는 4개의 비원어민 영어 억양(8명의 화자, 4개의 인도어) 화자의 음성 샘플이 포함되어 있습니다. 또한 4개의 영어 원어민 억양(4개국, 13명의 화자)과 대도시 인도 억양(2명의 화자)을 편집했습니다. 여기에서 사용할 수 있는 데이터 세트는 https://accentdb.github.io/#dataset 의 accentdb_extended 릴리스에 해당합니다.

나뉘다
'train' 17,313
  • 기능 구조 :
FeaturesDict({
    'audio': Audio(shape=(None,), dtype=int64),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=9),
    'speaker_id': string,
})
  • 기능 문서 :
특징 수업 모양 D타입 설명
풍모Dict
오디오 오디오 (없음,) int64
상표 클래스 레이블 int64
speaker_id 텐서
  • 인용 :
@InProceedings{ahamad-anand-bhargava:2020:LREC,
  author    = {Ahamad, Afroz  and  Anand, Ankit  and  Bhargava, Pranesh},
  title     = {AccentDB: A Database of Non-Native English Accents to Assist Neural Speech Recognition},
  booktitle      = {Proceedings of The 12th Language Resources and Evaluation Conference},
  month          = {May},
  year           = {2020},
  address        = {Marseille, France},
  publisher      = {European Language Resources Association},
  pages     = {5353--5360},
  url       = {https://www.aclweb.org/anthology/2020.lrec-1.659}
}