Prevents integer overflow UB in the shift rounding when the signal blows
up on bad bitstreams (if it triggers, the signal was already unusable
anyway).

Convert to 16 bits only at the very end

We were originally saturating the comb filter with enough margin
that the de-emphasis couldn't overflow at low frequency
(1/(1-.85) gain). The problem is that didn't leave enough headroom
for a full-scale HF signal.

Marcus Asteborg authored 4 years ago and Jean-Marc Valin committed 4 years ago

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

Reported in https://github.com/xiph/opus/issues/108

NaNs should be filtered at the Opus layer, so if there are any in the CELT
encoder, then it's likely something went horribly wrong (e.g. corrupted state).
In that case, better abort than have something bad happen.

Enables "safes" assertions even with ENABLE_ASSERTIONS isn't set

Cannot prove it's the correct value, but it's better than the previous
values, which sometimes segfaults. The increase was made necessary due
to 120 ms frame size support.

The change also makes the analysis run for sampling rates of 16 kHz and 24 kHz
since the features are only computed on the 0-8 kHz band. The longer time
window (20 ms instead of 10 ms) makes the tonality estimator more reliable
for low-pitch harmonics.

Speeds up encoding by another ~1-2%

Jonathan Lennox authored 9 years ago and Timothy B. Terriberry committed 8 years ago

Signed-off-by: Timothy B. Terriberry <tterribe@xiph.org>

Jonathan Lennox authored 9 years ago and Timothy B. Terriberry committed 8 years ago

Signed-off-by: Timothy B. Terriberry <tterribe@xiph.org>

We only keep when concealing less than 10ms with SILK.

gcc is better at optimizing it than the ?: version

Timothy B. Terriberry authored 11 years ago and Jean-Marc Valin committed 11 years ago

Optimizing celt_pitch_xcorr()/xcorr_kernel() which also speeds up
FIRs, IIRs and auto-correlations

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

Newer versions of MSVC are unhappy with the strategy of the build
 environment redefining "inline" (even though they don't support the
 actual keyword). Instead we define OPUS_INLINE to the right thing
 in opus_defines.h.

This is the same approach we use for restrict.

Code is still float, but at least tonality esitmation seems to work.
Speech/music analysis is still disabled.

Saves 2.5% on ARM without any asm. The float build still uses the old
code because celt_fir() then becomes a float function.