1dB Difference - an experiment in audibility

[STHLS5]

How many of us can positively identify a 1dB difference in your volume while listening to your favourite music? Maybe, users with digital volume with display could experiment themselves by setting your level to 85dB and then ask another person to increase or decrease the sound by 1dB.

Could you tell the difference? I am talking about the whole musical recording and not individual test tone. I can't. If you couldn't hear 1dB difference then how is it possible to hear a mere 0.5dB difference which mostly affect frequencies above 10kHz with different type of cables.

I would appreciate if someone could post a short recording with 1dB difference.

ST


+++++++++++++++++++++++++++++++++++++++++++++
{Moderator's comment: That's an interesting question. To make this a worthwhile test we need to think this through carefully. When you ask for a recording to demonstrate the audibility of a level difference (of one decibel) do you mean a 1dB difference across the whole audio band or in only some frequency bands? Please can I ask you to develop this idea further before calling for test material. Good experimentation anticipates the outcome and designs the test to prove or disprove the hypothesis. Over to you STHLS5 ....}

[STHLS5]

Let's choose a track, say Take Five. (1) In the first experiment let us listen at to the track at 85dB. Then another individual should either lower or increase or the volume by 1dB. He can also leave the volume at 85dB. The listener is then asked to guess if the volume is louder or softer or no changes.

(2) While the listener is listening another individual should increase or reduce the volume by 1dB. The listener (blindfolded) should correctly tell if there's a difference in the loudness.

(3) Ideally, if we can copy a song in the following manner [and across the entire audio spectrum]:

a) Original song without any adjustment.
b) The song is copied with 1dB less than the original level.
c) The song is copied with 1dB more than the original level.
d) The song is copied with 1dB difference randomly throughout the song.

I am not sure if it is possible to do (3) a,b and c with Nero's software.

I expect that no one here could positively detect the changes 100% accurately. If the listener can't detect the changes in the above sample. I doubt they could tell the difference of 0.5dB (cables effect) in the higher region of frequencies where there is so little musical content.

ST

[A.S.]

Drat. I typed a reply and went back a page in my browser and lost it. Again ....

A good experiment. Are you saying that we should create a post in which the visitor can play from the page a-d, with only a, the control (the reference) being identified?

You know, hearing acuity is the best researched area of human hearing with well over half a century of accrued knowledge. May I suggest that there are a couple of precursors which should be at least considered to get the best out of the experiment namely:

• Is 1dB a realistic level variation? It equates to a decibel change (sound pressure) of about 10%. What does the historical archive tell us about such a small change in loudness to the human ear?
• The selection of music - is Take 5 actually the best choice? It has an audio spectrum concentrated in the middle and upper range and as an old analogue recording from the 60s (or 50s?) is technically limited
• Should the listener be listening on headphones or speakers? It's not a trivial point. The earliest work on audibility was undertaken on listeners listening on mono headphones not to a loudspeaker and later work has shown that the headphone experience and speaker listening yield different results.

There are of course numerous other factors.

[STHLS5]

Is 1dB a realistic level variation? It equates to a decibel change (sound pressure) of about 10%. What does the historical archive tell us about such a small change in loudness to the human ear?

This is difficult for a layman like me to answer. From what I have read so far 1.0dB corresponds to the smallest change the our ears can detect. But other studies put the minimum discernable changes by our ears to be from 0.5dB to 3dB employing different methods.

On reflection, after considering what you are attempting to drive at, - maybe, the sample recording should contain three categories, i.e 0.5dB, 0.75db and 1db difference.

The selection of music - is Take 5 actually the best choice? It has an audio spectrum concentrated in the middle and upper range and as an old analogue recording from the 60s (or 50s?) is technically limited.

I have very limited collection of western music. I am unable to use my computer to do a spectrum analysis with the limited SACDs I have. But then, audiophiles claim to hear the difference in cables with vocal only tracks. So, I leave to you to choose a suitable composition for this experiment.

Should the listener be listening on headphones or speakers? It's not a trivial point. The earliest work on audibility was undertaken on listeners listening on mono headphones not to a loudspeaker and later work has shown that the headphone experience and speaker listening yield different results.

The whole exercise is to prove that a small difference of 0.5db at higher frequencies is impossible to discernible by us. The experiment is meant to be played over the loudspeakers.

Ideally, all the track samples are in wav and we should be able to burn them to CD to be played in our system. You are a better person to develop this idea into something meaningful.

ST

[honmanm]

One *might* be able to hear a 1dB level change with Alan's instant switchover box...

But I think that a change in some specific part of the spectrum stands more chance of being audible. Fortunately we all have an instrument to hand - speaker grilles. Do Harbeths sound different with their grilles off? And if so, what is the character of the change.

Once you've tried this, have a look for frequency response measurements for your speakers. e.g. Stereophile have a plot of the difference in frequency response of the P3ESR with and without grilles.

(personally I prefer grilles-on, which IIRC is the way they're supposed to be used)

{Moderator's comment: please do not use a screwdriver or other crude tool to remove the grilles from your speaker. An alternative test would be to drape a thin porous cloth over the tweeter ....}

[STHLS5]

One *might* be able to hear a 1dB level change with Alan's instant switchover box...

It may well be so. But can they identify whether any given track is definitely louder or softer, not just 'different'. Even after a five minutes break, you are unable to correctly tell if the track you are listening to is the higher or lower level sample as compared to what you heard yesterday.

So how could one able to pickup and compare sounds over a period of time?

ST

[A.S.]

...Even after a five minutes break, you are unable to correctly tell if the track you are listening to is the higher or lower level sample as compared to what you heard yesterday.

That strikes me as a truism - so blatantly obvious that it's a shame we have to repeat it here (but we do).

The human audio memory is incredibly short; so short that it has to be augmented by a method to sharpen the comparison between A and B i.e. to juxtapose event A and B into a timeframe in which the audio memory stands a chance of making a reliable, repeatable comparison - within one second, hence the switch-over-box.

I fully expect that should this archive be rediscovered again in fifty or a hundred years the most interesting curiosity to future researchers will be how so many 'audiophiles' at the beginning of the 21st century had a ludicrous self confidence in their audio memory when the tools to make instantaneous comparison were right at their finger tips, here. I don't and never have trusted my audio memory and it baffles me why anyone else would and then would potentially make a laughing stock of themselves be developing rigid and potentially erroneous opinions about the characteristics of A v. B which then pass into folklore.

But what do I know about it?

[A.S.]

About the audio test. It occurred to me overnight that (perhaps) in the interests of simplification, we could use not music but pink noise (hiss) as the test source. Why pink noise? Well, if you connect a CD player to a spectrum analyser and play a CD of typical classical music (not electronic pop music) and sample the whole CD you will arrive at a characteristic energy v. frequency statistical curve that (surprisingly?) doesn't vary much from music to music. The curve has a gentle tilt from low frequencies to high and means that on average, as frequency increases, there is less of it in music. That seems intuitively correct because bass frequencies are omni-present, and the highest frequencies (cymbal etc.) only appear occasionally so statistically their contribution is much smaller over the hour or so.

As it happens, pink noise has a characteristic low to high droop that's very similar to natural music, as opposed to white noise (FM inter-station hiss) which seems very bright and toppy because it has equal energy across the audio band, unlike music. So why don't we use pink noise as an initial test source and sidestep all the issues about selecting the optimum music? Then we can move on when we have gained some experience.

If you agree with my reasoning, here is an example of 15 seconds of pink noise, at a constant loudness in mono:





So that's what you'd hear if you sampled a CD every second for an hour, added together all the samples, and divided by 3600 (60secs x 60 mins.) to give you one seconds worth, then looped that 15 times to give the above clip. What do you think?

[STHLS5]

Thanks for the audio clip.

Pink noise maybe perfectly acceptable but audiophiles may say pink noise is not music and their golden ears were not trained to listen to noise. I was thinking about Time from Dark side of the moon. Just the first minute of so to avoid copyright issues.


ST

[A.S.]

Noted, but pink noise is the industry standard average music-like sound used for setting-up audio systems by ear, which is why those DIY room EQ boxes with supplied mic input use it to drive the speakers/room.

You yourself said that you didn't have a wide selection of western music. So we need to consider (which was very much in my mind when I proposed pink noise) that there is not only a spectral issue in selecting the optimum music to reveal any level difference but there is also a cultural one. As a crude approximation, western classical music is biased towards energy in the lower frequencies; oriental music is biased towards energy in the middle and upper frequencies.

What was your rationale when selecting the opening moments of Pink Floyd's Time? Are you sure that's a good choice? Apart from any other consideration, it is an analogue recording.

[STHLS5]

Time contains some high frequencies at the beginning and some reasonable bass. Mine is a SACD version so I couldn't play it in my PC. Meanwhile, I managed to figure out how to use Nero to adjust the volume at any spot across the spectrum. Here are 4 samples which were recorded at different loudness. Should I tell the level before hand? or should our readers try it by themselves and pick which sounds the best and the loudest. Maybe, a poll like survey be useful here.

ST

Four experimental loudness variation example MP3 files generated by STHLS5. Can you hear any differences between them?

Example X:



Example A:



Example Q:



Example Z:







{Moderator's comment: the problem with the Poll here on the HUG is that nobody bothers to vote. Second, the results are visible to all so they bias the vote.}

[STHLS5]

Thank you for the Flashplayer's clips.

Did you realize that just by changing the sequence of play from X,A,Q,Z to X, Q, A and Z your ears/brains perceive a different sample to be louder? The more I think about this experiment the more flaws of our senses revealed.

I spoke to some local hi-fi buddies who prefer musical samples rather than test tones. I am still waiting for someone to suggest a suitable composition. Meanwhile, I found this Decca recording that may meet our needs. Maybe, it is good to hear from someone who could positively hear sound of different cables to give a list CD or tracks.

BTW, if anyone wants to burn and try out the samples in the system. Please email me for wav samples. The wav files are 826KB each.
ST

{Moderator's comment: yes we did suspect that the order of pesentation influenced the results but we just followed your original order assuming that you had a reason to list that way. Now perhaps you appreciate why Alan was so keen to invest time up-front in eliminating as many variables as possible and to get a reall worthwhile result.}

[A.S.]

OK, I have not measured the levels this is my feedback based solely on what I hear on my plastic PC speakers ....

================================================== ========

X sounds the loudest to me.

A is of very similar or possibly slightly quieter than X.

Q is of markedly lower loudness. I'd guess that it was 3-6dB lower than X

Z definitely quieter than X, sounds of similar loudness to A

================================================== ========

How did I do?

[STHLS5]

..How did I do?

Excellent. A is minus 1dB. Q is -10dB and Z is minus 2dB. Did you get it right the first time or took many quick trials?

There is another aspect to consider. Should we use mp3 format? The file size for all three samples in wav format is 889,388 bytes. In mp3 it changed with different loudness.

ST

{Mod comment: you previous message is in the mod queue.}

[HUG-1]

We think MP3 is OK for this test. Yes there are some burbling artefacts but we are conscious of minimising the HUG database size as WAV (or MP3) binary files are not compressible in the backup archive. Text and images are.

[A.S.]

Did you get it right the first time or took many quick trials?

Actually I listened to the four pieces last night a couple of times, decided to wait until the morning and then replied after another two listens through. The notes I made last night after the first listening were similar to the final result I reported except last night I just couldn't decide if X and A were the same loudness or not. This morning I decided that possibly A was shade quieter than X but it was right on the edge of detectability (for me).

This says to me that when concentrating, listening in silence and when a virtually instantaneous comparison can be made it is barely possible to detect a 1dB difference in overall loudness. Even after toggling back and forwards repeatedly, the level difference is on the edge of detectability - for me anyway. If there was more than a second or two gap between the tracks and/or I looked away and/or some other external background noise and/or this level difference was less than 1dB then it would be completely impossible for me to reliably detect hear it. This result seems indicative of the motivated, interested, serious listener. For the casual disinterested listener, I'd expect a much greater difference would be needed.

[STHLS5]

I am not really happy about the intended results. It was quite obvious in my cheap about GBP5 PC speakers. Yes, I took several trials before coming up with the results. Whether the experiment is affected by the file conversion to SWF or the loudness level of the test or the closeness to the speakers affects the results are still unknown. This itself lead us to another experiment if the differences are more discernible in different format.

In the end, I am glad that even to Alan's most trained ears it was impossible to tell "Z" correctly as I expected.

================================================== ================================================== =
{Moderator's comment: If nothing else I hope that this experiment illustrates several important points to you which Alan already touched on and I list here: -

1] Spend ten times as long constructing the experiment, considering all the variables as actually doing the experiment
2] Do not expect the public to take any interest in and contribute to what to you is an interesting experiment. Past experience here confirms that. People will take, but few will give; that is the ethos of the internet
3] The public do not like to make fools of themselves especially when it concerns their main hobby
4] You have just one-shot at such an experiment. That chance has been used up.
5] We HUG management remain interested in such experiments. We are considering creating an exclusive membership controlled area to hold such advanced discussions but those who are permitted to join that select group would be expected to actively contribute to retain membership there.
6] Finally; 99.9% of audiophiles are entirely comfortable in their beliefs and have no interest or motivation to challenge them. The (impossible) art of the politician is to be inclusive to all even with very different and strongly held opinions.

[STHLS5]

I am still trying to grasp the difference in the file size of WAV and MP3 format. Re-reading the masking effect of in MP3 encoding it explains partially why MP3 files are smaller when the loudness is reduced quite significantly.

Strangely, the files in WAV remained the same. i.e 889,388 bytes. By looking at the files nothing is missing. This itself is interesting and take me one step closer to understand what information MP3 discards. I am asking myself the whether difference that I perceived in the earlier experiment here may have to do with the loudness level rather than the sound resolution or quality. Did I perceive the loudness to be better quality?

ST

[HUG-1]

Don't worry too much about the file size. If RLL coding is used in the data stream, 0000 of FFFF would be coded efficiently.

[STHLS5]

Agreed 100%. It doesn't matter if someone can show evidence that a US$3000 Lexicon BD-30 is nothing more than a re-badged US$1000 Oppo BDP-83 as seen in these pictures. Read more

And more re-badged brands here. There are consumers still willing to pay more for the same thing.

Someone reminded me that I must look into the whole system's synergy when I talk about cable effect. What can I say? It has been proven that a cheap US$700 system on a wooden stool sound as good as a more than US$12000 system. The test result showed that more people preferred the US$700 system. See http://www.matrixhifi.com/ENG_marco.htm .

It would have been nice if more people can share the same enthusiasm to get to the bottom of so many audiophiles myths.

ST