https://people.xiph.org/~jm/opus/opus-1.3/
describes the new features in the 1.3 release of the Opus codec,
including a new Ambisonics encoding method and an actual useful
usage of a neural network for deciding whether a given recording
is speech or audio, in realtime, in order to apply different audio
processing methods.
Includes demos of low-bitrate encoding for speech and mid, high
and uncompressed encoding for music.
At my desktop, using Sony MDR7506 headphones, I could reliably
tell the difference between the mid and uncompressed encoding for
all the samples (I think), but was not particularly reliable on
distinguishing the high rate and uncompressed encodings.
-dsr-