एक्सेंटडीबी

  • विवरण :

AccentDB संरचित और लेबल किए गए उच्चारण भाषण का एक बहु-जोड़ीदार समानांतर कॉर्पस है। इसमें अंग्रेजी के 4 गैर-देशी लहजे (8 वक्ताओं, 4 भारतीय भाषाओं) के बोलने वालों के भाषण के नमूने शामिल हैं; और अंग्रेजी के 4 मूल उच्चारण (4 देश, 13 वक्ता) और एक महानगरीय भारतीय उच्चारण (2 वक्ता) का संकलन भी है। यहां उपलब्ध डेटासेट https://accentdb.github.io/#dataset पर Accentdb_extended शीर्षक से रिलीज़ के अनुरूप है

विभाजित करना उदाहरण
'train' 17,313
  • फ़ीचर संरचना :
FeaturesDict({
    'audio': Audio(shape=(None,), dtype=int64),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=9),
    'speaker_id': string,
})
  • फ़ीचर दस्तावेज़ीकरण :
विशेषता कक्षा आकार डीटाइप विवरण
विशेषताएं डिक्ट
ऑडियो ऑडियो (कोई भी नहीं,) int64
लेबल क्लासलेबल int64
स्पीकर_आईडी टेन्सर डोरी
  • उद्धरण :
@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}
}