Diversifying Speech Audio Data

Diversifying Speech Audio Data

Researchers at the University of Essex have released a new speech dataset to study how people try to sound trustworthy. It includes approximately 1,000 audio clips from around 100 speakers, each recording sentences in both a neutral tone and a deliberately trustworthy tone. The clips are labeled with speaker demographics and vocal features, such as pitch and vocal clarity. Unlike existing datasets that rely on mostly white, younger speakers, thi…