สำเนียงdb

  • คำอธิบาย :

AccentDB เป็นคลังข้อมูลคู่ขนานแบบหลายคู่ของเสียงพูดเน้นเสียงที่มีโครงสร้างและมีป้ายกำกับ ประกอบด้วยตัวอย่างเสียงพูดจากผู้พูดภาษาอังกฤษ 4 สำเนียงที่ไม่ใช่เจ้าของภาษา (ผู้พูด 8 คน ภาษาอินเดีย 4 ภาษา); และยังมีการรวบรวมสำเนียงภาษาอังกฤษ 4 สำเนียงพื้นเมือง (4 ประเทศ 13 ลำโพง) และสำเนียงอินเดียนครหลวง (2 ลำโพง) ชุดข้อมูลที่มีอยู่ที่นี่สอดคล้องกับการเปิดตัวที่ชื่อว่า Accentdb_extended บน https://accentdb.github.io/#dataset

แยก ตัวอย่าง
'train' 17,313
  • โครงสร้างคุณลักษณะ :
FeaturesDict({
    'audio': Audio(shape=(None,), dtype=int64),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=9),
    'speaker_id': string,
})
  • เอกสารคุณสมบัติ :
ลักษณะเฉพาะ ระดับ รูปร่าง Dประเภท คำอธิบาย
คุณสมบัติDict
เสียง เครื่องเสียง (ไม่มี,) int64
ฉลาก ป้ายกำกับคลาส int64
รหัสลำโพง เทนเซอร์ สตริง
  • การอ้างอิง :
@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}
}