Project Relate: Breaking barriers with voice technology

Challenge

250M people worldwide have non-standard speech. For them, communicating with others and being easily understood can be difficult.

And the reality is that while India has a population of over a billion people, there are only about 4000 registered speech-language pathologists, audiologists and scientists.

That’s a big opportunity for speech technology. Unfortunately, standard speech recognition models don’t work very well for people with non standard speech. Each person’s speech patterns are very different, which makes it difficult for algorithms to understand and interpret them.

Following years of research from Google’s Speech and Research teams, Project Relate was created. An app that’s custom trained to unique speech patterns of people with non-standard speech.

Made possible by over a million speech samples recorded by participants of our research effort. Users can record multiple prompted speech samples, which is used in training a personalized speech model. These models are unique to the individuals who train them, and remain on device of the person.

The user can then use three features:
“Listen” transcribes speech in real time.
“Repeat” repeats what the speaker has said in a clear, computerized voice.
“Assistant” allows users to speak directly to their Google Assistant

Results

Through this project, Google hopes to help non-standard speech users be better understood on a daily basis and make products that rely on speech recognition technology, like Google Assistant, more accessible to everyone.

The app has been piloted with English users in India and will be expanded to Hindi users in 2023.