This dataset contains Swahili speech audio paired with transcriptions.It is split into training, development, and test sets with no speaker overlap.
Split
Samples
Hours
train
60,216
359.98
dev
3,258
19.30
dev_test
3,117
18.62
Total
66,591
397.89
id – unique sample ID
audio – audio file
text – transcription
prompt – prompt text