 jm
|
9314bad169
Re-arranged the wideband mode so that programs using narrowband only and
|
пре 17 година |
|
jm
|
bdcefb47d2
Put all the OS-dependent calls in os_support.h
|
пре 17 година |
|
jm
|
2c15c034ee
Fixed a DTX bug by changing the order of the test in case we have a null mode.
|
пре 17 година |
|
jm
|
ad6f7d03ef
bit of cleaning up
|
пре 17 година |
|
jm
|
38cbb2c7e4
Fixed-point: vocoder conversion complete
|
пре 17 година |
|
jm
|
cc79e5b334
Replaced speex_error() by speex_fatal() and speex_assert()
|
пре 17 година |
|
jm
|
b39b09af76
Fixed-point: converting some of the vocoder code
|
пре 17 година |
|
jm
|
e485ebe5bf
Tuned the vocoder gains and prevented the ol_gain from being reduced by
|
пре 17 година |
|
jm
|
5c8ab69180
fixed valgrind support (and gitignore)
|
пре 17 година |
|
jm
|
0f6335ea9e
Implemented SPEEX_GET_ACTIVITY
|
пре 17 година |
|
jm
|
b292d62669
Replaced some warnings by notifications when it's not necessarily caused
|
пре 17 година |
|
jm
|
80b3286613
pseudo-stack is now shared between the narrowband and wideband layers.
|
пре 17 година |
|
jm
|
d92e9bd2bf
Doing some sanity checking (just in case) on the propagated long-term
|
пре 17 година |
|
jm
|
e1ee270db3
Fixed minor pi_gain synchronisation problem in the wideband decoder. Should
|
пре 17 година |
|
jm
|
08151da50f
Only the RMS of the excitation is saved in the encoder -- 320 bytes off the
|
пре 18 година |
|
jm
|
576d051794
simplified the double_codebook code and improved accuracy slightly
|
пре 18 година |
|
jm
|
da7f92f056
code shuffling -- no real change.
|
пре 18 година |
|
jm
|
123ad25d33
fixed-point: minor tweaks
|
пре 18 година |
|
jm
|
06235982ef
fixed-point: wideband double codebook converted
|
пре 18 година |
|
jm
|
71d983adf9
Removed 5 kB of useless buffering in the wideband encoder. Also, fixed the
|
пре 18 година |
|
jm
|
32accf86c4
Got rid of the excitation memory in the decoder and changed the relevant vars
|
пре 18 година |
|
jm
|
304e9007d3
reviving the old EPIC_48K mode.
|
пре 18 година |
|
jm
|
b544a9278b
Saved innovation from nb to wb is now 16-bit precision.
|
пре 18 година |
|
jm
|
9c84ddb524
More debug info and fixed two unlikely overflows on clipped input.
|
пре 18 година |
|
jm
|
a702231023
Removed the generic PSHR/SHL/SHR operators and changed them to either the
|
пре 18 година |
|
jm
|
95861b7310
Fixed an overflow in the excitation decoding (again on heavily-clipped signal)
|
пре 18 година |
|
jm
|
d4695483ae
Third (hopefully last) set of fixes for the overflow problem on clipped input.
|
пре 18 година |
|
jm
|
3a2222946e
Adding the "f" suffix to float constants (path by Michael Jerris).
|
пре 18 година |
|
jm
|
8157ec116a
more replacements of int with spx_int32_t
|
пре 18 година |
|
jm
|
86bb320ffb
Changed part of the API to use spx_int32_t instead of int (works better on
|
пре 18 година |
