A Better Audio Comparison Tool
I've mentioned a few times the difficulty in using the software tool, DiffMaker - and others have experienced similar issues. So, I'm thinking seriously about attempting to put together a "better" utility, something that will be more intelligent in correlating between different versions of sampled waveforms, wherever they came from. And it will be specifically aimed at trying to ferret out the deep down differences, not the obvious stuff like frequency response, phase shifts and the like - but rather the low level, detail variations that seem to be so important in subjective quality evaluation.
To give an idea of the "challenge", consider taking 2 recordings from 2 vinyl plays of a particular record. Having already seen what results, this is impossible for the likes of DiffMaker to handle - the constant speed variation between the 2 captures completely defeats it. But this what I would aim to be able to deal with, to synchronise in meaningful ways between the 2 versions and then report on what is interesting in the effective difference ...
Update, 14/7: Still chuggin' - the tools and their functionality look pretty good; still working, slowly, :), while learning to use the tools well, on easy/smart ways to get the versions synchronised. This is the real meat of the deal, a far bit of fiddling required to get this happening nicely ...
To give an idea of the "challenge", consider taking 2 recordings from 2 vinyl plays of a particular record. Having already seen what results, this is impossible for the likes of DiffMaker to handle - the constant speed variation between the 2 captures completely defeats it. But this what I would aim to be able to deal with, to synchronise in meaningful ways between the 2 versions and then report on what is interesting in the effective difference ...
Update, 14/7: Still chuggin' - the tools and their functionality look pretty good; still working, slowly, :), while learning to use the tools well, on easy/smart ways to get the versions synchronised. This is the real meat of the deal, a far bit of fiddling required to get this happening nicely ...
Total Comments 5
Comments
-
Sounds like fun Frank - a piece of software! My suggestion would be to make it a toolbox of stuff, rather than try to do everything automatically like Diffmaker tries to do. A human guiding the process would be much more useful.
I had an idea about testing a DAC by implementing a digital adaptive filter on one DAC (the reference one, multibit of course) and then seeing how good a null could be created when the other's an S-D DAC.... The adaptive filter could take account of the FR errors.Posted 3rd June 2014 at 12:41 AM by abraxalito
-
Yes, I feel software has more overall potential, because with hardware there is always the issue of whether the hardware tasked with the comparing is compromising the performances of the DUTs. With software, once an accurate as possible capture is made then you're free to analyse to your heart's content.
I agree with the interactivity, a closed box approach as DiffMaker does is a dead end; one should be able to partially analyse, with certain settings; and then adjust the settings as one goes "deeper", as a type of filter, to focus on a particular area.
To kick off, I'll try using standard software tools that are out there, combine them to see if they have enough processing horsepower, as is, to get something meaty out. If progress is made then do a standalone version of that ... and see where that goes ... :)Posted 3rd June 2014 at 10:30 PM by fas42
Updated 3rd June 2014 at 10:32 PM by fas42 -
Okay, having a bit of a thrash with octave, the freebie version of Matlab - the first time I've played with this type of software. Looks very interesting, lots of potential here - the usual, really irritating software glitches - the plotting module wouldn't behave itself on my laptop, yet was fine on the desktop; so had to figure out a workaround to make it useful, and convenient ... have I mentioned that I hate software, :D ...?Posted 10th June 2014 at 11:33 PM by fas42
-
Octave is looking very good - still chuggin' along, feeling my way into the language - which is what it is, Yet Another Tool for playing with numbers. It's demonstrating a very good ability for performing comparison exercises - I've got 2 versions of a music track - which are at different speeds, therefore start and stop times don't nicely mesh - currently working on approaches to get the two to synchronise in meaningful ways ...Posted 19th June 2014 at 01:04 AM by fas42
-
I[I] have[/I] picked an interesting pair of tracks to use as a test run: very poor quality in the recording sense, one from a CD, the other an extract of fairly high resolution audio from a YouTube video - are they the same track ...??
No, as mentioned above they run at different speeds, and further, the speed alters during the track - taking the CD as reference, the AAC version changes speed as it plays! The latter appears not to be vinyl, and in fact is quieter in a noise sense than the CD, looking at the waveform, at high frequencies! Even more 'laughable' - the spectrum of the YouTube version shows the intrinsic above 10kHz band is "better", there's more there! To listen to them at a casual level they sound very similar - but visually there is substantial, obvious variance between the waveforms. Which is not to say that the compressed version has "lost" information in an obvious sense - every single wriggle of the CD version is perfectly matched by a wriggle in the YouTube variant, just that the wriggles don't beautifully match, they won't "subtract" neatly.
Interesting ...Posted 24th June 2014 at 01:46 AM by fas42
