-
Notifications
You must be signed in to change notification settings - Fork 1
Datasets
DASS (see Kretzschmar et al. 2013) is an audio corpus that records 64 interviews (3-4 hours each) of Southern speeches featuring dialects in eight Southern states with a mixture of ethnicities, ages, social classes, and education levels. DASS provides fruitful resources for researchers to work on. It would be interesting to see how well the model trained on general North American English corpora performs on these Southern speeches. The audio data is also publicly accessible from http://www.lap.uga.edu/, and it is also available via the Linguistic Data Consortium (https://catalog.ldc.upenn.edu/LDC2016S05).
We used LibriSpeech ASR corpus as our major training dataset. LibriSpeech is a corpus of approximately 1000 hours of 16kHz read English speech, prepared by Vassil Panayotov with the assistance of Daniel Povey. To access, go to http://openslr.org/resources/12.
Kretzschmar, W. A., Paulina B., Jacqueline H., Lee P., Ilkka J., Lisa L.O., and Tapio S. (2013). "The Digital Archive of Southern Speech (DASS)." Southern Journal of Linguistics, 27 (2). 17–38.