[opus] Antw: Re: 133 kbps stereo killer sample
jmvalin at jmvalin.ca
Tue Apr 18 21:53:30 UTC 2017
On 18/04/17 03:15 AM, Ulrich Windl wrote:
> Somewhere I read the recommendation that the peak level of input material
> shopuld not be 0dB, but something like -3dB. The reasoning was that encoding
> reconstructs the audio signal from the sampling points, and the reconsruction
> could actually exceed the sampling point, which my result in an overdrive,
> which will result in clipping.
> I just wonder whether the encoder (or the decoder at least) can or should warn
> in such situations.
The sad truth is that most recent music is actually mastered to be as
loud as possible (see "loudness war"). So warning wouldn't be useful
since we would just end up warning all the time. There is some
decoder-side mitigation in libopus to reduce the artefacts on clipping,
but unfortunately, they are not currently supported by opusdec.
>> Opus remains unbeaten for me at 133 kbps. That's totally awesome.
>> On 10 April 2017 at 17:14, Agustín Dall'Alba <agustin at dallalba.com.ar>
>>> Hello! I found a sample I can ABX successfully when encoded at
>>> 133.333 kbps. I was targetting 1 MB/min.
>>> floex.wav is 4:54–5:04 of a lossless copy of 'Forget-me-not' by
>>> Floex, downloaded from http://store.floex.cz/album/zorya
>>> floex-133.opus was created with `opusenc --bitrate 133.333333 floex.wav
>>> floex-133.opus`, opus-tools 0.1.9 (using libopus 1.1.4).
>>> I used a Sony MDR-V6 pair of headphones and my laptop's oonboard audio. I
>>> detected the artifact when I was listening casually from my phone with the
>>> same headphones.
>>> In the second to last note of the crescendo (at ~5 s in the sample)
>>> a slight flutter in the original that turns more prominent and tape-like
>>> after encoding.
>>> I also tried encoding at 128 kbps and 150 kbps.
>>> ABX results:
>>> - floex-133.opus.wav 10/10, p-value<0.001
>>> - floex-150.opus.wav 4/10, p-value<0.83
>>> - floex-128.opus.wav 17/20, p-value<0.002
>>> Only at 150 kbps is Opus transparent to me for this sample.
>> opus mailing list
>> opus at xiph.org
> opus mailing list
> opus at xiph.org
More information about the opus