 jm
|
123ad25d33
fixed-point: minor tweaks
|
18 éve |
|
jm
|
408e18edac
No more calls to log() in the wideband encoder.
|
18 éve |
|
jm
|
ae72157da4
fixed-point: wideband folding gain quantization as a codebook (instead of log)
|
18 éve |
|
jm
|
ac023fc5ec
some comments and a minor tweak
|
18 éve |
|
jm
|
06235982ef
fixed-point: wideband double codebook converted
|
18 éve |
|
jm
|
2ff77b3f37
wideband excitation folding (including gain) now in fixed-point.
|
18 éve |
|
jm
|
b933a2a51d
more wideband fixed-point work (folding_gain is now in fixed-point).
|
18 éve |
|
jm
|
4f64d9545d
more wideband fixed-point work
|
18 éve |
|
jm
|
c3d7a3f302
wideband fixed-point details
|
18 éve |
|
jm
|
6b539b654e
Fixed a very rare overflow problem in the wideband mode, while slightly
|
18 éve |
|
jm
|
fa592b75b3
Making use of the wideband encoder input buffer to save memory: 640 bytes off
|
18 éve |
|
jm
|
809a4507f4
Most of the wideband encoder now in 16-bit. Also, saved another 1.5 kB off the
|
18 éve |
|
jm
|
a7cb0d33d4
zero-response now in 16-bit and stored on the stack. About 1 kB saved off the
|
18 éve |
|
jm
|
8f2809edc8
No need for an explicit buffer to receive the narrowband innovation.
|
18 éve |
|
jm
|
71d983adf9
Removed 5 kB of useless buffering in the wideband encoder. Also, fixed the
|
18 éve |
|
jm
|
678e1bd9ce
Another 1 kB off the encoder state by converting the qmf_decomp output to
|
18 éve |
|
jm
|
32accf86c4
Got rid of the excitation memory in the decoder and changed the relevant vars
|
18 éve |
|
jm
|
b544a9278b
Saved innovation from nb to wb is now 16-bit precision.
|
18 éve |
|
jm
|
c679c22021
another 640 bytes off the decoder state.
|
18 éve |
|
jm
|
29dba66464
Saved another 2kB from the decoder by converting high-band synthesis to
|
18 éve |
|
jm
|
a629adae51
Replaced fir_mem_up by qmf_synth which doesn't require intermediate
|
18 éve |
|
jm
|
9d2371d0fc
oops. EXTEND32() wasn't needed here.
|
18 éve |
|
jm
|
a702231023
Removed the generic PSHR/SHL/SHR operators and changed them to either the
|
18 éve |
|
jm
|
95861b7310
Fixed an overflow in the excitation decoding (again on heavily-clipped signal)
|
18 éve |
|
jm
|
3a2222946e
Adding the "f" suffix to float constants (path by Michael Jerris).
|
18 éve |
|
jm
|
8157ec116a
more replacements of int with spx_int32_t
|
18 éve |
|
jm
|
86bb320ffb
Changed part of the API to use spx_int32_t instead of int (works better on
|
18 éve |
|
jm
|
56b9080347
forgot to free memory for low_innov. sb_celp is now 100% leak free (AFAIK)
|
18 éve |
|
jm
|
cfd022ecb3
Making libspeex C++-friendly
|
18 éve |
|
jm
|
697521cfb7
cleaning up FIXMEs
|
18 éve |
