Re: Questions on Opus encoding

On Tuesday, August 12, 2014 06:19:09 PM Ulrich Windl wrote:
> Hello!
> 
> I hope this is the right audience for Opus codec questions:
> From what I've read about Opus I got the impression that the codec will
> switch between speech and music encoding rather quickly. I wanted to try
> that for a CD track that has a vocal (talk) introduction followed by music
> (just two instruments). Using foobar2000 with some custom setup for Opus, I
> tried VBR encoding with a 256kbps setting (opus-tools-0.1.9-win32). My
> expectation was that the bitrate would be significantly lower than 256kbps
> while speech is encoded, while it would go up for the music part. However
> during playback I could not see a significant difference in foobar's
> (version 1.3.2 before you ask) display. Is this the way things should be?
> The other thing that confused me is that foobar displays 48kHz for the
> sampling rate in file properties (a bug?). Opusinfo displays "Original
> sample rate: 44100Hz".
[...]
As I understand it, one should think of the "bitrate" not as an actual 
"bitrate" but more like a "quality" setting that just happens to go from 0 to 
510 and coincidentally happens to roughly track the the resulting bitrate...

As far as the "speech" encoding, that (again, as I understand it) really
only comes into play at lower bitrates where the tradeoff in detail vs bitrate 
seems worth it (at higher bitrates, there are plenty of "bits" available to 
get the details of speech sounds with the high-quality general-purpose 
encoding method, so the special speech-optimized mode never kicks in.)

From what I remember, it's roughly 20kb to 32kb where the hybrid 
"speech/music" mode starts kicking in when the encoder thinks it's
advantageous to do so, and down below that where it tends to go 
all-speech-mode (if the encoder thinks the input is probably
mostly speech).

Above that the bit savings from including the special speech mode
isn't worth the tradeoff in quality (you could just reduce the bitrate/quality
setting that much and still end up with higher-quality results).

Subjectively for me, it seems like there's really no point in encoding opus 
above about 128kbps at most for 2-channel audio. In the default 96kbps to
128kbps range the sound quality is already top-quality to my ears, and doesn't
really leave more than negligible room to hear any more improvement no matter 
how many bits you give it to use (from my personal perspective, I think if 
you're willing to throw 256+kbps at stereo audio, you might as well switch
over to FLAC instead.)

(I have no experience with >2-channel audio, so while I assume more bits would 
be useful there I can't really make any predictions).

tl;dr: the special speech mode is for preserving speech quality at lower 
(<32kbps) bitrates rather than preserving bits at higher bitrates, and you
can probably chop the file you're encoding down to half or less of what you're 
using and probably not notice any difference in sound quality with opus.