-
WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Paper: Digital Compared To Analog - Listening Test
- Thread starter Wombat
- Start date
What a strange experimental protocol.
watchnerd
Grand Contributor
I thought this thread was about digital paper vs analog paper.
So disappointed....
So disappointed....
- Joined
- Feb 23, 2016
- Messages
- 20,767
- Likes
- 37,626
So is digital paper composed of just bits.........of paper?
watchnerd
Grand Contributor
Don't be ridiculous, that's papier mache.
"Digital paper", designed to "be paper":
- Joined
- Feb 23, 2016
- Messages
- 20,767
- Likes
- 37,626
Yes, and even stranger results, and even stranger still conclusions of those results.
So we have two paths to make reproduced sound more like unprocessed analog in surround:
(A) Use low sample rate digital with high bandwidth amps and speakers. It sounds more analog than high sample rates over the same gear.
or
(B), use high sample rate recordings with bandwidth limited microphones, speakers and amplifiers. As it sounds more analog than lower rate digital over the same gear.
I wonder if it is possible to garner whether condition (A) or condition (B) is closest to analog sound?
I suppose a combination of aliasing and IMD would actually fit those conditions, and make this seem a logical result.
Condition (B) is the easiest to implement. Most everyone already has bandwidth limited mics, and speakers while also having high sample rate DACs.
I'm not knowledgeable enough to really say, but their statistical modeling seemed unusual to me. I'm not sure what to make of that.
Guermantes
Senior Member
What a coincidence! I was reading this paper just yesterday, however the copy I have (via AES) includes a critical response that appeared in the JAES from Sylvain Choisel (also at Philips) who questions the assumptions that the analysis relies on, along with a reply by the original authors (JAES Vol.57 No.10, 2009 October).
In their response, the original authors seem to feel that their study is somewhat tentative and conclude with the following:
"Considering that the experience of professional recording and mastering engineers working with high-resolution audio has not been confirmed and quantified in laboratory tests, we do not have a good validation of the testing methods used in our industry. The current subjective testing
methodology does not seem to sufficiently reveal or amplify the features characterizing individual listening experiences. Perhaps this methodology, which is derived from food and fragrance testing, is not as readily effective for investigating subtle auditory sensations and the experience of music? We too would like to encourage more work in this area."
In their response, the original authors seem to feel that their study is somewhat tentative and conclude with the following:
"Considering that the experience of professional recording and mastering engineers working with high-resolution audio has not been confirmed and quantified in laboratory tests, we do not have a good validation of the testing methods used in our industry. The current subjective testing
methodology does not seem to sufficiently reveal or amplify the features characterizing individual listening experiences. Perhaps this methodology, which is derived from food and fragrance testing, is not as readily effective for investigating subtle auditory sensations and the experience of music? We too would like to encourage more work in this area."
watchnerd
Grand Contributor
I'm baffled....
Why was the reference point the analog system and not the live performance they just heard?
And if the analog system was the reference, why have the live portion at all?
Confused.
watchnerd
Grand Contributor
Or maybe it's a simple "people like compression" -- take your pick where you do it, either in the electronics or in the software?
- Joined
- Feb 23, 2016
- Messages
- 20,767
- Likes
- 37,626
But that doesn't really fit the results.
- Joined
- Feb 23, 2016
- Messages
- 20,767
- Likes
- 37,626
One of the things about food and fragrance testing is they usually start with a difference that is clearly perceptible. If they did the same thing with audio it should maybe restrict bandwidth to 3 khz and move up in steps. And it usually involves something more defined like which is sweeter or which has the same sweetness as reference. Which is more analog sounding is not so clear in what the testees were trying to perceive. Yes they had an analog reference, but what is analog sound, and what is the continuum of perception of analog-ness? Asking about bass response, or treble response is more like what food and fragrance testing does.
watchnerd
Grand Contributor
I can't make sense of their results....so I'm reaching...
- Joined
- Feb 23, 2016
- Messages
- 20,767
- Likes
- 37,626
I think test takers never heard the live system really live. They were calling the real time analog playback "live".
- Joined
- Feb 23, 2016
- Messages
- 20,767
- Likes
- 37,626
Okay understood.
I think I summarized their conclusions with conditions A and B correctly about which results were more analog like.
Now I wonder about some of the statistical hoky poky to determine what is a p=.05 result. The math part is fine, but the assumptions about what would be expected seem like an over-reach to me. If the model's assumptions are wrong then so is the rest. But I don't have any expertise in the area. Perhaps to someone who does I am foolishly misguided on the modeling.
I think assumption 1 is way too much to assume.
Assumption 1 There is a one-dimensional latent (nonmeasured) variable that denotes the perceived quality of the audio systems. This variable determines the outcomes of the listing test for the subjects [3].
They then use the listening test results based upon assumptions 1 and 2 (two seems reasonable to me: that the AD/DA chain cannot sound better than straight analog feed) to create a model of expected results. In fact, I missed this on the first reading, they didn't get results that indicate p=.05 confidence. The overall combined results are within range to what you'd expect if the one dimensional latent variable is not perceived.
They then go on to show under the two conditions wide bandwidth playback and 20 khz playback the results tend to skew slightly in opposite directions though still not to the point of meeting p=.05 for either test conditions C1 or C2 considered individually. But they base their conclusions on this minor difference. That high sample rates are heard more like analog, but the high frequency response of such rates is not the reason why.
Looking at this again, the conclusions aren't supported. And they are curious as if they were self serving for someone's preconceptions.
I really don't think these results show anything.
Theo
Active Member
- Joined
- Mar 31, 2018
- Messages
- 288
- Likes
- 182
It seems that they had a unique MC reproduction system (apparently good quality) fed alternately with 3 different sources : direct "live", "red book" sampled, hi-res sampled. No recording involved. So, the conclusion I get from this is that their hi res sampling equipment was distorting the signal
and that the standard sampling could not be differentiated from the direct path. So, not good for hi resSimilar threads
