[opus] Impossible two bugs in Opus

[Emily Bowman]

 On Fri, Nov 16, 2018 at 7:40 PM ongaku zettai <sergeinakamoto at gmail.com>
wrote:

> Hello.
> i have over 30GB of Opus music and noticed that
> solo instrumentals and solo vocals uses more bitrate
> than full-mixes.
> Here's example where Opus 1.3 used 190 kbps for
> piano solo and 159 kbps for full-mix.
> (--bitrate 160 --music)
> Download example piano solo 15MB:
> https://mega.nz/#!wLBz3AZT!YmqQMkAGqc4kGHumNWZAfB7Cmcf4vFlHpT6IiiAVCNA
> FLAC uses 2 times less bitrate for solo than full-mix because
> it contain less data, naturally.
> Hence i think it's a bug in Opus.
>

Why would you expect FLAC's and Opus's wildly different methods of
compression to be comparable? You won't understand anything about a lossy
codec by comparing it to FLAC. Whether a piece has "more data" or not only
depends on what measure of entropy you use; in Opus's case, the encoder
believes the solo has less entropy because it can't throw as much away
while maintaining quality, even though the full mix has less entropy if you
retain every bit.


On Sat, Nov 17, 2018 at 5:37 AM ongaku zettai <sergeinakamoto at gmail.com>
wrote:

> Hello. Me again.
> Have you tried to encode piano solo?
> Noticed high bitrate Opus gave?
> And there's also artefact at 15kHz
> which wasn't in the original audio.
> Visible with Spek program.
> Download FLAC and Opus both files,
> new link:
> http://www.filedropper.com/example_3
> FLAC full: 1084 kbps;
> FLAC solo: 465 kbps.
> with --bitrate 160:
> Opus full: 158 kbps;
> Opus solo: 190 kbps.
> Included also Spek spectrogram PNG file
> for Opus solo to see the artefacts.
> Music is copyrighted but we using it
> for testing purposes, ok?
>

There's no point in even looking at it. A spectrogram isn't going to tell
you anything useful about whether an audible artifact exists, and is mostly
useless except in lossless codecs, unless you're just plain curious about
what gets masked to give you the end result. The whole point of
psychoacoustic masking is to remove frequencies that can't be heard, please
don't call it a bug. If there's an audible artifact, that's much more
interesting.

Is this your first foray into lossy codecs? You might want to familiarize
yourself with the behavior of the other modern standards, like MP3, AAC,
and Vorbis, as well, instead of singling out Opus without understanding its
peers. (MP3 in particular is much simpler to understand, if you want to
follow the details, or even MP2, since there's no spectral replication or
noise or anything like that.) hydrogenaudio.io is a good place to start.

Em
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.xiph.org/pipermail/opus/attachments/20181117/ee33698e/attachment.html>