Calls are deferred to the actual loss and we only process the minimum
required.

Jan Buethe authored 2 years ago and Jean-Marc Valin committed 2 years ago

Model now stores LPC gamma, look-ahead, and end-to-end.
Parameters aren't quite reliable yet, YMMV

Packet loss concealment based on LPCNet (work in progress)

Krishna Subramani authored 3 years ago and Jean-Marc Valin committed 3 years ago

Making LPC computation and prediction differentiable

Jean-Marc Valin authored 3 years ago and Jean-Marc Valin committed 3 years ago

Down to 20 features

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

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

Would cause unused pitch embedding vectors to be used