Sound Quality Vs. Measurements

[dvv]

Originally Posted by fas42

Just to remind myself, I tried it again. I took a track from the 80's, what many people would regard as pop fluff, Dexy's Midnight Runners - Come on Eileen, which started as a 128kbps MP3, size of just under 4 Mbytes, 4 1/4 mins running time. Imported into Audacity, resampled to 384000 rate, exported as 24 bit WAV file. Resampling on Dell laptop took 1/2 hr with CPU at 90% usage, exporting took another 1/4 hr, final file size 571 Mbytes!!

But this silk purse project works, for me. Once the DAC had warmed up a bit, the difference between original and resampled was chalk and cheese. The MP3 is a scratchy, unpleasant mess, the hi res conversion had good tonality of the instruments, nice space rendition, depth, imaging, all the usual adjectives.

Sometimes, miracles are possible ... . Just wondering, whether anyone else has tried this bizarre exercise ...?

Frank

Thank you, Frank.

Personally, I haven't tried anything like it myself, because I see no sense or possible gain in upsampling a signal which has already had a part of it thrown away, as MP3 does. "Thrown away" means forever lost.

The difference in sound quality you note I'd attribute to better DAC operation, because I don't see any other possible answer (which is not to say there is none).

[dvv]

Originally Posted by abraxalito

No, I disagree - your example was an example of brand politics/marketing rather than engineering optimisation.

You cited that the compromise was made for 'sales optimum' - that's just a number. A number is without value - hence this was done because bean counters had to be placated. Separating the engineering design from the marketing would lead to the same result - compromise, and probably this split personality effect is present in abundance in GM's culture.

Disagree all you like, but it is a fact that in this case, the system is optimal if it satisfies the output goal, which is sales. And it does.

Despite the fact that the product is suboptimal technically. If it were the other way around, it would have been technically optimal and too expensive to compete, and would hence be suboptimal overall.

As for sales figures being just "numbers", well, you should try to explain that to the thousands of people working to make it happen, I feel relative sure that to them, it's a hell of a lot more than just a "number". To them, it's their job, their livelyhood. Also, I'd point out that since Chevrolet is only one of GM's divisions, these "numbers" also have influence on the macrosystem called GM as much more than just "numbers".

The only thing I agree with you is that there is much internal politics in deciding what is the optimal performance of this system; unfortunately, GM has not one but TWO divisions operating on the same markets, and are thus competing with themselves, like it or not. This does impose certain constraints on making purely engineering decisions, to be sure.

[tsiros]

Originally Posted by DF96

You certainly don't get an output signal which looks/measures anything like the input signal.

what kind of observation would you make and what kind of measurement would you take?

because i i don't think that you can tell between uncompressed and 320 Kbps mp3

[abraxalito]

Originally Posted by dvv

Disagree all you like, but it is a fact that in this case, the system is optimal if it satisfies the output goal, which is sales. And it does.

Then we must mean different things by 'optimal'. Fair enough.

As for sales figures being just "numbers", well, you should try to explain that to the thousands of people working to make it happen, I feel relative sure that to them, it's a hell of a lot more than just a "number". To them, it's their job, their livelyhood.

Well a true optimisation would take account of these people's happiness too, unlike your compromise solution.

The only thing I agree with you is that there is much internal politics in deciding what is the optimal performance of this system

More evidence that our meanings of 'optimal' differ - mine is not subject to opinion like yours.

; unfortunately, GM has not one but TWO divisions operating on the same markets, and are thus competing with themselves, like it or not.

Why is that 'unfortunate' ? Do GM's senior management see it the same way?

This does impose certain constraints on making purely engineering decisions, to be sure.

Which sounds as though you are now agreeing with me

[nigel pearson]

When looking at music on an oscilloscope years before digital I was always struck by how simple the waves looked in comparison to the complex sounds I heard . I guess we detect start and stop points and we also recreate the sounds we hear from information clues ? I was told that speech was almost impossible for computers to comprehended in the early days . Not a question of clear speaking . The computer could not detect words . The gaps we hear so clearly are not apparent . I can reasonably cope in French . I find French films more difficult than real life especially after digital . I watched a film called Life is just one long tranquil river ( my translation and not precise ) . I understood it better than usual . Then I twigged it , Lille , I learned my French in Tournai some 9 miles away . I also learned in the south . Parisian French , not a chance . Great film .

https://en.wikipedia.org/wiki/Life_I...ng_Quiet_River

BTW , big river .

https://www.youtube.com/watch?v=5AWC...eature=related

[vacuphile]

Originally Posted by nigel pearson

When looking at music on an oscilloscope years before digital I was always struck by how simple the waves looked in comparison to the complex sounds I heard .

The ear does an 'FFT' on the incoming sound and what the brain sees is a landscape of different bin intensities. Nothing like what you see on the scope.

vac

[nigel pearson]

Originally Posted by dvv

Thank you, Frank.

Personally, I haven't tried anything like it myself, because I see no sense or possible gain in upsampling a signal which has already had a part of it thrown away, as MP3 does. "Thrown away" means forever lost.

The difference in sound quality you note I'd attribute to better DAC operation, because I don't see any other possible answer (which is not to say there is none).

If I am right that about start and stop points then perhaps that is what is being better defined ? The ear I was told ( Oxford University ) seemed to be a very low quality analogue device with digital back up . A very fast signal returns from the brain which they had no idea as to it's function . I suggested a servo loop . I supplied a pair of selected KEF T27 with above 40 kHz ability for the tests .

http://www.jneurosci.org/content/19/19/8704.full.pdf

[fas42]

Looks like I'm going to have to repeat some of the experiments I did a year or 2 ago. I was playing with LAME, attempting to squeeze the last ounce of quality out of encoding to MP3, and I was quite impressed with what could be achieved -- there were subtle differences but you had to really concentrate to be aware of the variations. Plus, I seem to remember I could get the difference file to almost 60dB down, does that seem reasonable?

What I might do is grab a very well recorded hi res file, encode with LAME as best I can to 320, and 128 rates, then reverse the process in Audacity as I described in the earlier posts, and finally listen to all versions on the PC speakers. Will be interesting to see what comes up ...

Also, would be fascinating if someone else, please, tried doing this and what they come up with ...

Frank

[nigel pearson]

Originally Posted by vacuphile

The ear does an 'FFT' on the incoming sound and what the brain sees is a landscape of different bin intensities. Nothing like what you see on the scope.

vac

Yep .

[abraxalito]

Originally Posted by fas42

Also, would be fascinating if someone else, please, tried doing this and what they come up with ...

Have done it, with 320k only. No surprises - most of the time the null is better than 60dB if I remember correctly, often lower than 70dB. It depends on the complexity of the music. mp3 works pretty much as advertised.