Google has released a new automatic speech recognition app, dubbed Live Transcribe.
The application is aimed at helping people who have significant hearing loss or are profoundly deaf. According to the World Health Organisation 466 million people experience hearing loss; that’s roughly five percent of the world’s population.
Once the Android application is open, it uses a phone’s inbuilt microphone to listen for spoken word, then transcribes the audio in real-time onto the screen.
To make this happen without guzzling data or suffering latency issues as it connects to the cloud, it uses an on-device neural network-based speech detector, which automates network connections to a cloud-based automatic speech recognition (ASR) engine.
(This builds on work Google has carried out on AudioSet, its collection of over two million 10 second YouTube excerpts, open sourced on Github in 2017).
Tested by Computer Business Review, it performed highly accurately.
“Live Transcribe has the potential to give people who are deaf or hard of hearing greater independence in their everyday interactions,” Google’s Brian Kemler said.
One of the key research scientists on the project, Dimitri Kanevsky, has been deaf since an early age. Kemler said he typically uses services such as Communication Access Realtime Translation (CART). Not only is this a costly way around the problem, it also involves a high level of organisation before any meetings take place.
The project was born when one of Kanevsky’s colleagues saw the difficulties he was having with meetings and general interactions.
That led Google engineer Chet Gnegy to start working with Google’s accessibility team to build a product that would help. Currently in its beta release, the Live Transcribe application supports 70 languages and dialects.
While the interface and setting in the application are stripped back, it still contains some clever features that will benefit users with profound hearing disabilities, such as a vibrate function that alerts the user when a speaker begins talking again after more than 10 seconds of silence.
“It also enables two-way conversation via a type-back keyboard for users who can’t or don’t want to speak, and connects with external microphones to improve transcription accuracy,” comments Brian Kemler. If the application is to be used by children you can even set it to mask any profane language it detects, which again worked ******* nicely when tested by Computer Business Review.
The Live Transcribe application is currently available for download in its beta form in the Google Play store.