Our technology was initially designed to be used with external ASR. For sentences within our 18 domains, where the ASR is correct, we see accuracy (defined by m

Our technology was initially designed to be used with external ASR. For sentences within our 18 domains, where the ASR is correct, we see accuracy (defined by mapping into the current domain and action) around 90%. The reality of course though is ASR is sometimes wrong, which does decrease our accuracy in practice. The best direct demonstration of practical accuracy is just try out the app ;)