When navigating by voice, traditional street addresses can be difficult to input. what3words cuts out the frustration, getting people easily and precisely from A to B.pic.twitter.com/ho7B0EPeUK
-
Interested to see whether the
#ASR fails more often when reciting 3 random words that don’t make contextual sense together. May particularly be a problem with accents. Many ASR systems rely on a Language Model to resolve beam search to wreck a nice beach.
Hi Justin, we have programmed the speech recognition system (Cerence) to hear three random words as a what3words address rather than trying to build a sentence or street address from them. We have also accounted for accents. This gives accuracy of above 95%.
