Masterclassing Webinars

what3words and Speechmatics launch speech recognition and post-processing API

David Murphy

Location tech firm what3words has released an end-to-end speech recognition and post-processing API, in conjunction with speech recognistion and machine learning firm, Speechmatics.

The what3words Voice API allows a user to say three words into any application or service, to have it return the what3words address and corresponding GPS coordinates, all through a single API call. The Speechmatics integration provides speech recognition and machine learning capabilities that enable new what3words languages to be built on demand. The companies said this will unlock the potential of voice devices for more people across the globe.

what3words has divided the world into a grid of 3-metre squares, and given each square a unique address made of three words. It enables people to identify and share any precise location using just three words. what3words is the first addressing system optimised for voice input, eliminating many of the frustrations encountered when using voice to enter traditional street addresses.

Speaking a street address into a voice interface, such as an in-car navigation system, can be lengthy and frustrating. Saying ‘Takeme to 241st Street’ can sound exactly the same as ‘Take me to 2, 41st Street’. Street addresses also use thousands of non-dictionary words, the pronunciation of which can be near impossible to guess. The town of Godmanchester, for example, is actually pronounced ‘Gumster’.  Each what3words address is unique and similar what3words addresses are placed far apart, making user errors easy to detect and correct using the what3words AutoSuggest post-processing function.

For customers that want to quickly enable what3words-only voice search, this single API removes the commercial and technical complexity of fusing together multiple APIs, meaning developers can be up and running in a matter of hours. With the product hosted by servers across numerous key international hubs, the audio stream’s latency is low enough to respond to the user in real time.

“The what3words Voice API in partnership with Speechmatics means that what3words address searches by voice can now all be done in this single product.” This makes it the ideal solution for apps and services that want to enable simple address entry via voice, without the speech recognition ambiguity associated with address search through traditional ASR systems – duplicate names, for example.”

The what3words Voice API powered by Speechmatics currently supports English, Mandarin, Hindi, Japanese, Korean, German, Arabic and Spanish.

YOU MAY ALSO BE INTERESTED IN