This shows you the differences between two versions of the page.
private:low_latency_asr_for_the_deaf [2018/07/05 17:12] (current)
|Line 1:||Line 1:|
|+||===== Low latency ASR for the deaf =====|
|+||This has roots in "the colour of speech" and en-to-end models on characters or phonetic spellings.|
|+||People process words. They are very used to reading and for clean input we can do low latency ASR that is only a few words behind.|
|+||DECISION: Abandon "the colour of speech" and en-to-end models. These won't have good word breaks and so will take a long time to learn. They will be overtaken by standard ASR.|
|+||DECISION: Don't do standard ASR as it competes with Speechmatics.|