Meta has created a dataset to improve accuracy in speech recognition AI systems. The dataset contains audio recordings and transcriptions of over 27,000 verbal commands and messages from around 600 people, as well as self-reported demographic information. The commands and messages followed prompts relevant to voice assistant tools, including notification controls, dictation, calling, messaging, music, photo or video capture, and utilities. Meta organized the dataset by clustering commands or messages of similar content together to enable researchers to improve the accuracy of speech recognition systems across different demographics.
Image credit: Flickr user drestwn
