This release covers 14 coarse-grained intents from the Banking domain. The dataset is split into:
test-100samples per intenttrain->650samples per intent
Audios were generated by 11 (Indian English) speakers, recording over a telephony line. Also includes anonymised speaker information - like native language, languages spoken, native language and gender - to allow more structured discussions around robustness and bias.