Google noticed that the transcriptions Voice provided for voicemail weren’t particularly accurate. More often than not, they were nearly indecipherable. Fortunately, Google has decided to do something about that, and new and improved transcription is on the way to Google Voice.
The improved voicemail system for Google Voice was announced in a post on Google’s official blog yesterday. The new system is powered by a long short-term memory deep recurrent neural network, which Google says cuts errors in transcription by up to 49 percent.
As part of the work on the new system, Google asked users to provide their voicemails. “Thanks to those who participated, we are happy to announce an improved voicemail system in Google Voice and Project Fi that delivers more accurate transcriptions,” Google software engineer Zander Danko wrote in the blog post.
While Google Voice users will no doubt be happy to have any improvement in the service’s voicemail transcription, the real reason we’re seeing this is hinted at above. Google’s pay-as-you-go wireless plan Project Fi uses a lot of technology from the Google Voice project, and Google likely doesn’t want the service to launch with transcription being as bad as it currently is.
This isn’t the first time Google Voice and Project Fi have been mentioned in the same breath, though it’s better news this time around. In May, invitations to Project Fi began rolling out, but Google Voice users who signed up found their numbers automatically pulled into the service, whether they planned to port it or not. Google fixed this shortly afterward, but it was a harrowing few days for some customers.
To use Google Voice’s new transcription, you don’t have to do anything special — just keep using the service and you’ll hopefully notice that the transcription is suddenly good, or at the very least less bad.
