Remove old pitch estimator and retrain all models

Totally untested -- most likely doesn't work

Untested

saves a few cycles

And missing prototypes

Should be able to handle all previous GRU variants and more.

Marcus Asteborg authored 1 year ago and Jean-Marc Valin committed 1 year ago

Signed-off-by: Jean-Marc Valin <jmvalin@amazon.com>

first attempt of C implementation of fec encoder (not tested yet due to NEON/DOT_PROD not being separable)

Only on the C side, no sparse GRU B training yet

Adapted from PR: https://github.com/mozilla/LPCNet/pull/134
by zhuxiaoxu <zhuxiaoxu@ainirobot.com>
but had to be reworked due to previous weight quantization changes.

Avoids having to compute a sigmoid

It no longer overwrites its input vector

Saves on the MDense/softmax computation since we only need to compute
8 values instead of 256.

Using rational function approximation for tanh() and sigmoid.