Context-aware Communication Aids (CoCA)

About the project

Speech recognition technology has developed rapidly in recent years. However, these advances risk leaving behind those who would find it of greatest value: people whose speech is difficult for others to understand. To address this, CoCA aims to codevelop with end users digital communication aids that meet their needs and aspirations.

Over the last 20 years the use of speech technology has grown rapidly. Nowadays, without a second thought people can ask their phone a question ("Siri: what time is it?") or tell a smart speaker to play them some music ("Alexa: play the Rolling Stones").

Or at least that’s the idea. Unfortunately, at the moment these systems do not work well for people whose speech differs from what the technology expects. So, for example, people with strong regional accents, people speaking a second language and people whose speech is difficult for others to understand may all find that these systems simply don’t work for them.

In other words, today’s speech technology is not accessible to everyone. And for many people who find communication difficult, the value of access to these technologies could go far beyond simple interactions with Siri or Alexa. For example, speech recognition systems could be quicker and more convenient than scan-and-switch interfaces when making notes or interacting with other people, while smart speakers can act as interfaces to environmental controls as well as entertainment systems.

The use of speech technology is only going to increase in the coming years and, as researchers, we would like to ensure it becomes accessible to all. Our long-term objective is the development of machine learning-based “context-aware communication aids” for people whose speech is difficult for others to understand.

Initial work has focused on:

• Investigating with potential users of such aids of their opinions and how they might participate in the research needed to develop them.

• Developing and beta-testing of a prototype web-based service for collecting speech data of a nature considered likely to be necessary to develop the aids.