HydrogenAudio

Hydrogenaudio Forum => Scientific Discussion => Topic started by: Serge Smirnoff on 2018-08-21 13:02:41

Title: Objective evaluation of SBC and aptX codecs using music-based audio metric
Post by: Serge Smirnoff on 2018-08-21 13:02:41
I made objective measurements of the codecs using music-based df-metric. The main purpose of the post - to present the method. Your feedback is valuable as always. Thanks in advance.

The article is here - http://soundexpert.org/news/-/blogs/audio-quality-of-bluetooth-aptx#aptxmeasurements

Concerning the results, aptX is no better than SBC, pure marketing.
Title: Re: Objective evaluation of SBC and aptX codecs using music-based audio metric
Post by: Big_Berny on 2018-08-21 14:26:29
Well the main problem of your test setting is that it's not "objective" as you think. A bigger difference in wave forms doesn't necesserily mean that the difference to the original sound file is more audible. So aptX and SBC(loudness) may look similar in your analysis but one might outperform the other in listening tests.

So a listening test as mentioned here would lead to much more valid results: https://wiki.hydrogenaud.io/index.php?title=ABC/HR
Title: Re: Objective evaluation of SBC and aptX codecs using music-based audio metric
Post by: Serge Smirnoff on 2018-08-21 20:39:55
Well the main problem of your test setting is that it's not "objective" as you think.
Objective in my post means measurable (not subjective).
Quote
A bigger difference in wave forms doesn't necesserily mean that the difference to the original sound file is more audible. So aptX and SBC(loudness) may look similar in your analysis but one might outperform the other in listening tests.
Your "doesn't necessarily" is correct. It means that in some cases the relation holds, in some - doesn't. And this is the subject of my main research - to discover in what cases bigger (measurable) differences in waveforms necessarily lead to their better audibility, and in what cases such relation is absent. Proposed method of artifacts' comparison aimed at discovering such useful cases when the size of differences in waveforms correlate well to results of listening tests.

At this stage of the research I use the following selection criteria for the useful cases: if distances between artifact signatures are less than 0.3 within a group of devices under test, then their perceived sound quality can be judged by the size of their waveform differences.

The above criteria I used in this small SBC/aptX research.
Title: Re: Objective evaluation of SBC and aptX codecs using music-based audio metric
Post by: Big_Berny on 2018-08-22 13:58:39
Well the main problem of your test setting is that it's not "objective" as you think.
Objective in my post means measurable (not subjective).
True, it's objective. What I meant was that external validity is very limited with this test scenario.

Thanks for clarifying the other points btw. Sounds interesting.