Building better speech recognition for everyone

Speech recognition technologies are used in a wide variety of commercial products, including voice search, voice typing, voice translation, automatic captioning, and more. They work well – almost at human level – for standard varieties of some larger languages like English or French. These technologies are even used in professional settings where a high degree of accuracy is required: to help medical professionals to take notes, to provide captioning for videos and recordings, and to allow travelers and the international business community to communicate in other languages.

However, when it comes to many other languages and non-standard language varieties, the availability and quality of speech recognition systems is generally poorer. For many languages, including large languages of Africa, India, and Indonesia, these technologies are not yet available, and various reports have demonstrated that recognition quality is lower for speakers with different sociolects, dialects, and non-native accents.

In this talk, we discuss the problem of building speech recognition systems for accented speakers and new languages, covering various topics from data gathering to system evaluation that might help to create better speech recognition for everyone.

Bio: 

Alëna is a linguist at Google, where she works on topics related to speech processing, automatic speech recognition,

and model evaluation pipelines. Prior to joining Google Speech, she focused on formal language theory and its applications.

She received her Ph.D. in Computational Linguistics from Stony Brook University. She was a recipient of the Junior

Researcher Award from Stony Brook University's Institute of Advanced Computational Sciences.

*ATTENDANCE IS WELCOME IN PERSON OR ONLINE*

Join Zoom Meeting

https://stonybrook.zoom.us/j/93876616016?pwd=NlhyV2kweXA4dVduR2U3VldjaFY4Zz09

Meeting ID: 938 7661 6016

Passcode: 888323