Speech Technology Magazine

Webinar -  Everybody's Talking: The Path to Conversational Customer Care

Register now for this September 26, 2017 Event!

The 2017 Speech Industry Luminaries: Tom Ash

By The Editors of Speech Technology - Posted Aug 26, 2017
Page1 of 1
Bookmark and Share

THE SPEECH TRANSCRIBER 

Tom Ash

speech recognition director at Speechmatics

Prior to 2016, Speechmatics, an automatic speech recognition technology provider based in Cambridge, England, had made very effective products for the English language but struggled mightily to get similar performance with other languages. The big breakthrough came earlier this year when the company launched a unique machine learning technology and artificial intelligence framework that allows it to add almost any language automatically.

“This is an exciting time to be in speech recognition,” Tony Robinson, Speechmatics’ founder and owner, said in a statement at the time. “We are at the forefront of how deep neural networks are changing speech recognition. With our ever-expanding and highly experienced R&D team, we continue to push the boundaries in speech technology, especially around languages, accuracy, and deployment.”

As speech recognition director at Speechmatics, Tom Ash can take a lot of the credit.

After launching the AI framework, called Auto-Auto, in mid-January, Speechmatics was able to roll out new languages very quickly, typically at a pace of one every two weeks or so. At press time, the company had rolled out 29 languages, including German, Spanish, and Russian.

The secret was to have as little manual intervention as possible for each language once the original data was in place. To accomplish that, Ash had to change the company mind-set first, moving from a manual to an automated approach. Then he implemented a single pipeline to run the system so nothing was fractured into separate specialties. The final step in the process was creating the framework that allows developers to use previous language builds to inform future ones, even when limited data is available.

In so doing, Ash developed speech recognition technology that not only provides significantly greater accuracy but further expands its commercial applicability for call compliance, content discovery, and speech analytics, among other use cases.

Ash joined Speechmatics in 2012, shortly after earning his doctorate in neuroscience at the University of Cambridge. He also was a programming and computational modeling consultant for a year.

“Without the work from Tom and his team, Speechmatics would not be at the forefront of ASR or competing with the biggest technology giants in the world,” says Georgina Robertson, director of marketing at Speechmatics.

Page1 of 1