 jm
|
01f90f0703
Fixed a major regression in the 4 kbps mode by adding the target update to
|
%!s(int64=18) %!d(string=hai) anos |
|
jm
|
ec0c8224b1
Got rid of ~64 divides per sub-frame and saved 512 bytes of scratch space.
|
%!s(int64=18) %!d(string=hai) anos |
|
jm
|
b19fc2b3c5
minor RAM tweak: delayed stack allocation to reduce max scratch memory use
|
%!s(int64=18) %!d(string=hai) anos |
|
jm
|
db6c7f3850
weighted signal now in 16-bit precision, saving 500 bytes in the encoder state
|
%!s(int64=18) %!d(string=hai) anos |
|
jm
|
c7ba68d11a
oops. Fixed an off-by-one that could (at least theoretically) cause stack
|
%!s(int64=18) %!d(string=hai) anos |
|
jm
|
ce3aa230f8
cleanup and saved 480 bytes of RAM in the decoder
|
%!s(int64=18) %!d(string=hai) anos |
|
jm
|
e9b37e7937
bit of cleaning up. More to come.
|
%!s(int64=18) %!d(string=hai) anos |
|
jm
|
248f06d8cf
Getting rid of 25600 divide operations/sec in the pitch search.
|
%!s(int64=18) %!d(string=hai) anos |
|
jm
|
ac766f9ea8
new experimental comb filter code
|
%!s(int64=18) %!d(string=hai) anos |
|
jm
|
101a3c067c
fixed potential problems with float rounding causing NaNs
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
0fc7fe6c0f
Added MAX16 operator
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
7aa63e8176
Blackfin assembly for compute_pitch_error
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
e5ab0eac33
Separated time-critical portion of the pitch gain computation so it can easily
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
3f26eeb943
Removed unnecessary call to open_loop_nbest_pitch (which in turn fixes
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
b5bc6a8e3e
Packet loss concealment converted to fixed-point. Added hooks for blackfin
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
48a7c60e4f
Cleaned up arch-dependent optimizations
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
61619db166
removed some // comments
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
eb475e2e2b
Added some control on the aggressiveness of the pitch predictor in the form of
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
e92f830694
Think I fixed the PLC slowdown due to denorm/underflow. Also don't re-
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
99951f31f1
convert codebook data (signed char) to spx_word16_t in a cleaner manner
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
3e814b072a
Now autodetects (and handles) size of integer types.
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
0ece9021de
some cleaning up
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
6fba5c1ba0
More work on fixed-point operators
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
0e21c0dd91
Fixed-point improvements (moved some stuff to 16-bit arithmetic)
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
716239f876
Removed another bunch of warnings (when using some of the -W options)
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
ea52aea0da
fixed-point cleanup, removed some warnings
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
6f0e96a431
Merged a modified version of Jamey Hicks' C55 patch, fixed a
|
%!s(int64=19) %!d(string=hai) anos |
|
jm
|
bc55d853b7
Now possible to put temporary arrays directly on the (real) stack
|
%!s(int64=20) %!d(string=hai) anos |
|
jm
|
c02fbf4301
cleanup
|
%!s(int64=20) %!d(string=hai) anos |
|
jm
|
88fe6a1cc0
oops, that should fix the float version
|
%!s(int64=20) %!d(string=hai) anos |
