Hello,
I am trying to build a DeepSpeech lib for the mobile device and I found the high usage of the CPU. In the codes, I found Eigen is used in KenLM to decode.
The question is that how could I confirm NEON is activated for Eigen in decoding? If it is not, what should I do in compiling to open it?
I’d really appreciate any guidance. Thank you!
“high usage” without context is not actionable ; proper neon speedup is in place, as long as you are running something compatible.
For example, I found the usage of CPU on iPhone 7plus is 150% when running deepspeech recognition in my App. Could I get some log to check neon is opening?
By the way, could I think that neon is also activated for eign in KenLM when I use the language model to decode inference resuls ?
Thanks a lot!