The initialiser already takes care of this

Using the min energy of the two last non-transient frames rather
than the min of just the two last frames. Also slightly increasing
the "thresh" upper bound coefficient to 0.5.

di_max was counting the _number_ of code-points remaining, not the
 largest one that could be used.

By moving the energy floor to the encoder, we can use a different
floor for prediction than for the decay level. Also, the fixed-point
dynamic range has been increased to avoid overflows when a fixed-point
decoder is used on a stream encoded in floating-point.

If someone messed up scaling, they'll know because it'll sound bad

The most important thing is to not screw up the energy. Then we
compare the bit-rates.

This was computing on uninitialized data (though the result was never
used)

Gregory Maxwell authored 14 years ago and Jean-Marc Valin committed 14 years ago

Correct an encoder/decoder mismatch at low volume levels. Relax some low level clamps so that the dynamic range can extend further below the 16bit floor.

Also a fix for the zero-ing of unused band energies.

Timothy B. Terriberry authored 14 years ago and Jean-Marc Valin committed 14 years ago

We were always storing collapse_masks[i*C+1], which could have
 overflowed the buffer on the last band in mono.
This also moves the stores outside the conditional, which allows
 common code to be used for masking the high bits, address
 generation, etc.

This provides more entropy and allows some more flexibility on the
encoder side.

Timothy B. Terriberry authored 14 years ago and Jean-Marc Valin committed 14 years ago

Jean-Marc's original anti-collapse patch used a threshold on the
 content of a decoded band to determine whether or not it should
 be filled with random noise.
Since this is highly sensitive to the accuracy of the
 implementation, it could lead to significant decoder output
 differences even if decoding error up to that point was relatively
 small.

This patch detects collapsed bands from the output of the vector
 quantizer, using exact integer arithmetic.
It makes two simplifying assumptions:
 a) If either input to haar1() is non-zero during TF resolution
     adjustments, then the output will be non-zero.
 b) If the content of a block is non-zero in any of the bands that
     are used for folding, then the folded output will be non-zero.
b) in particular is likely to be false when SPREAD_NONE is used.
It also ignores the case where mid and side are orthogonal in
 stereo_merge, but this is relatively unlikely.
This misses just over 3% of the cases that Jean-Marc's anti-collapse
 detection strategy would catch, but does not mis-classify any (all
 detected collapses are true collapses).

This patch overloads the "fill" parameter to mark which blocks have
 non-zero content for folding.
As a consequence, if a set of blocks on one side of a split has
 collapsed, _no_ folding is done: the result would be zero anyway,
 except for short blocks with SPREAD_AGGRESSIVE that are split down
 to a single block, but a) that means a lot of bits were available
 so a collapse is unlikely and b) anti-collapse can fill the block
 anyway, if it's used.
This also means that if itheta==0 or itheta==16384, we no longer
 fold at all on that side (even with long blocks), since we'd be
 multiplying the result by zero anyway.

This looks for bands in each short block that have no energy. For
each of these "collapsed" bands, noise is injected to have an
energy equal to the minimum of the two previous frames for that band.
The mechanism can be used whenever there are 4 or more MDCTs (otherwise
no complete collapse is possible) and is signalled with one bit just
before the final fine energy bits.

Based on spreading_decision()'s logic. We choose tapsets
with less roll-off when we think the HF are tonal.

Supporting three different tapsets with different roll-offs. The default
is now a 5-tap post-filter with a 13 kHz cutoff frequency.

Gregory Maxwell authored 14 years ago and Jean-Marc Valin committed 14 years ago

In CVBR mode the rate selection was failing to add bytes which were about to fall off the end of the bitres and never be reusable, causing undershoot.

In case start or end changes, we want the encoder and decoder
to be in sync and not do anything stupid.

Nor scaling the mid only after we've been able to store it
for folding.

Timothy B. Terriberry authored 14 years ago and Jean-Marc Valin committed 14 years ago

Adds a new bitexact_log2tan() function which is much simpler, and
 more accurate.
The new approximation has an RMS error of 0.0038 bits from the
 correctly rounded result over the range of inputs we use, compared
 to an RMS error of 0.013 for the old log2_frac() method.
The actual computation of delta is also changed to use FRAC_MUL16,
 since this allows us to keep the full accuracy of the new method
 while avoiding 16-bit overflow.
The old delta computation actually could overflow 16 bits: it needed
 8 for the log2_frac() result, 1 for the sign of the difference, and
 8 more for N.

Also removes the 8 byte/packet lower bound