Starting with compute_linear()

Saves ~270 kB of weights in the decoder

Removing some indirections

Only code non-degenerate symbols, which makes the encoder faster

Felicia Lim authored 1 year ago and Jean-Marc Valin committed 1 year ago

Update silk/x86/NSQ_del_dec_sse4_1.c to match the remaining
silk/NSQ_del_dec.c changes made in
c913dc38



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

Felicia Lim authored 1 year ago and Jean-Marc Valin committed 1 year ago

Co-authored-by: James Zern <jzern@google.com>

Avoids switching to CNG unless we just have a "refresh packet"

Including removing useless code

Removing one of the 2d conv layers for pitch estimation reduces
complexity without noticeable degradation. FARGAN model has more
adversarial training.
Also, no need for the double precision in the low-pass filter.

The sinc filter offset was incorrectly handled. Since it perfectly compensates
for the analysis offset, nothing has to be done.
Also, the preemphasis memory was never initialized.

Gets us about 2x speedup on x86