The encoder would reserve bytes for redundancy and then decide to use
CELT, but not use those bytes.

Also avoids stupidly high/low rates

Also clarify in doc that conformance issue relates to decoder only.

The multistream encoder input has st->layout.nb_channels channels.
The number of channels actually encoded is st->layout.nb_streams +
st->layout.nb_coupled_streams, which may be fewer, so ideally it
would only analyze those when computing the optimal frame size, but
there is no code to do that currently.

Divide by the actual number of channels mixed

Adding a "channels" argument to compute_frame_energy() and is_digital_silence()

This avoids an overflow on extreme signals

Thanks to Mark Harris for pointing it out.

Gives another ~2% speedup

Speeds up the encoder by about 1%

Felicia Lim authored 8 years ago and Timothy B. Terriberry committed 8 years ago

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

That was likely a mistake from a misunderstanding on the SILK bit allocation

Hybrid CBR now simply forces the SILK CBR rate to the "target" value. Also,
we're getting rid of the -2 kb/s offset for CBR, which appeared to be harmful.
In the case of hybrid constrained VBR, the cap is computed in the same way
as the target.

Previously, the bandwidth detection could (e.g.) change the bandwidth
from fullband to superwideband, and the hysteresis would then cause
bandwidth to be stuck in superwideband.

Felicia Lim authored 8 years ago and Jean-Marc Valin committed 8 years ago

Computes a pseudo SNR as the ratio between the running peak signal
energy to instantaneous noise energy. DTX is turned on only if this is
higher than a pre-determined threshold.

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

Felicia Lim authored 8 years ago and Jean-Marc Valin committed 8 years ago

A frame is marked as valid for DTX if it contains noise or only digital
silence. As before, there is an overhang period of 200 ms and a maximum
consecutive DTX period of 400 ms.

If the new DTX cannot be used because of the complexity setting and
sampling frequency chosen, the SILK DTX will be used instead.

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

Felicia Lim authored 8 years ago and Jean-Marc Valin committed 8 years ago

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

Should be able to make better mode/bandwidth decisions when CBR/complexity
are involved.

The new attenuation is much smaller than previously.

New MLP doesn't attempt to classify silence as speech/music

Gradual downmix now happens in the 24 kb/s to 32 kb/s range

No values outside of +/-1 detected now.