stanford-oval/genie-cloud

Add streaming STT API

gcampax opened this issue · 0 comments

Streaming STT has lower-latency than REST STT, and it also provides live-feedback, so in practice we always use it, by talking directly to the MS speech service, which exposes the MS key in the app. That creates security problems (key abuse) and operational problems (key rotation).
Instead, we should wrap the streaming STT API, which would let us deal with all the ugliness of the MS speech API using the SDK, and also do speech + NLU at the same time, for lower latency.