On February 4, 2019, Google released Live Transcribe, an Android app that turns the speech around a user into live captions on the phone screen, so people who are deaf or hard of hearing can follow a face-to-face conversation in real time. At launch it supported more than 70 languages and dialects, covering over 80 percent of the world’s population.
Technically, the app pairs cloud-based speech recognition with an on-device neural network that detects when speech is present, which lets it run continuously without burning through mobile data on silence. The team made a deliberate interface choice to omit per-word confidence scores after research showed they distracted users without adding conversational value. Google built the app in close partnership with Gallaudet University, the deaf-serving institution, whose researchers helped design and validate it. One of its core developers, Dimitri Kanevsky, is himself deaf and also worked on Google’s Project Euphonia. Live Transcribe shipped pre-installed on Pixel 3 phones and rolled out on the Play Store.
Why business readers should care: Live Transcribe shows the payoff of co-designing a product with the community it serves, rather than for them. The decision to hide confidence scores - counterintuitive to engineers - came directly from user research, and it is the kind of detail that determines whether an accessibility tool is actually adopted.