Accelerators support?
Opened this issue · 4 comments
claudio-xx commented
Is it possible to support accelerators (like Coral TPU, or a GPU)?
I run Whisper on an i7 but it takes 10s or more to recognize a command (like "spegni luce veranda", using tiny-int8), and error rate is quite high with smaller models.
AlexxIT commented
I haven't seen a working implementation
alex-savin commented
Is it possible to use that project as an example?
https://github.com/google-coral/project-keyword-spotter
AlexxIT commented
No. Project completely different and not related to Whisper
alex-savin commented
@AlexxIT you are right, it could be useful only for the openwakeword