As we round the corner to 2022’s final stretch, there’s a great deal that team Speechmatics can be proud of. We’ve picked up a number of awards, including Boldest AI (and we’re nominated for a hatful more, so watch this space). We raised $62m in Series B Funding and introduced new features like Entity Formatting and Language Identification to our product offering.
Perhaps for me, one of the moments I’ll look back on with the greatest pride – in this calendar year – was our introduction of 14 new languages, bringing our total to 50. While that number itself is impressive, it’s what it means as a milestone that matters: Speechmatics now supports language coverage for over half the world’s population.
Setting and Achieving Goals
When I came to Speechmatics as CEO, the team’s immense drive made an immediate impression on me. From engineering through to marketing and sales, ours is the sort of team who, once they’ve put their mind to a goal, won’t stop until it’s been accomplished. It’s because of this forward-thinking, goal-orientated mindset, that we’ve got to the point of enabling over 50% of the world to access speech-to-text technology. You can find the full list of the 14 new languages here.
Another thing about this accomplishment which impresses me, is the inclusionary nature of some of the languages chosen to pursue. It makes commercial sense to add Vietnamese, a language spoken by over 87 million, but in choosing to support less commonly spoken languages like Welsh and Basque (less than a million native speakers each), we’re living by our word when it comes to our aim of understanding every voice. We know what those languages mean to the communities who speak them and we want to ensure they’re represented.
Not Just a Number
While 50 is an impressive marker, there’s an argument that we could simply provide an engine offering many more languages to a passable standard. But that’s not the Speechmatics approach. We’ll never tick boxes simply to say we have. If we offer a new language, we insist it’s of the highest accuracy. This is why our Enhanced model has been trained to a higher standard than Google* across our new shared languages. To give just one example from our recent release, our accuracy for Estonian is 85.95%, compared to Google’s 67.21%**.
It’s thanks to our insistence on innovation – like our advancements in self-supervised learning –which has led to an overall higher success rate in speech-to-text for all voices.
More to Come
Making sure every voice is heard has always been written into the DNA of Speechmatics. As John Hughes, our Accuracy Team Lead said of the accomplishment, “Our aim has always been to understand every voice and so it’s vital that we also capture languages that may not be as well recognised. This has allowed us to provide the most comprehensive offering on the market.”
As for the future, my amazing team tell me we’ve got even more languages in the pipeline, with an aim to reach 70% global coverage by 2025. As ever, we won’t release any languages until they’re tried, tested, and fit for purpose. We’ll continue marching forward in the search for full global coverage.
Katy Wigdahl, CEO, Speechmatics
*Speechmatics and Google have eight of the new languages in common: Basque, Estonian, Galician, Marathi, Mongolian, Tamil, Vietnamese, and Thai.
**Accuracy is measured as the percentage value of 1−WER (Word Error Rate, the rate at which words are incorrectly inserted, deleted or substituted).