[Ethan Winer]

I don't want to take the Capacitors burn-in thread off topic, so I figured it's best to start a new thread. This is a great topic for The Lab because it addresses what types of changes can be heard, and what types are too small to worry about. Separating fact from fancy makes us more knowledgeable, and in turn smarter consumers.

The conventional wisdom is that 1 dB is about the smallest change in level or frequency response that people can reliably identify. This is also known in science as the Just Noticeable Difference, or JND. However, the JND for level and response changes depends on the frequencies present, as well as their absolute volume level. At loud volumes the JND is smaller than at very soft levels. Likewise, JND is smaller at frequencies where our hearing is most sensitive. In other words, we can identify smaller level changes when the music is loud enough to hear clearly. So under ideal conditions - when listening at a decent level in a room with no damaging reflections, or when using earphones - it's possible to identify level and response differences smaller than 1 dB.

The book, Introduction to the psychology of hearing by Brian C. J. Moore states that the smallest detectable volume change for broadband noise is between 0.5 and 1.0 dB. Section 19.2.1 of Sound Reproduction: The Acoustics and Psychoacoustics of Loudspeakers and Rooms by Floyd Toole shows that changes in frequency response as small as 1/3 dB can be heard under ideal circumstances. Both of these are scholarly works that refer to real research, versus the 0.03 dB JND claimed anecdotally with no further explanation by a hi-fi magazine writer linked in the other thread.

However, when discussing changes in frequency response, versus volume level, two additional factors come into play. First, and perhaps too obvious, the frequencies being boosted or cut must be present in source. You could boost 50 Hz by 20 dB on a recorded flute solo, but that won't be noticed because flutes contain no content at such a low frequency. If such a boost is heard, it's surely due to other content in the recording, such as air conditioner rumble or other noises unrelated to the flute itself.

Another factor is the bandwidth of the boost or cut. When dealing with broadband content such as a full music mix or pink noise, the JND for a boost or cut with a wide bandwidth (low Q) is smaller than when only a narrow range of frequencies is changed (high Q). This too makes sense, because more total energy is affected with a wide bandwidth. In Toole's book linked above, his JND of 1/3 dB is for a boost having Q of 1. When the Q is raised to 10 the JND rises to 3 dB before the response change can be noticed by listeners.

--Ethan

[stereocilia]

Of course the rise/fall time and the intensity make  differences, too.

Back in the day, audiologists used a test called SISI.  I can't remember exactly what that stands for, something like – short increment sensitivity index.  Anyway, it involves a pure tone presented at (I think) 20 dB above threshold.  Every few seconds it increases by 1 dB and then goes back to baseline.  If the patient can reliably detect these blips, then this is a sign of cochlear pathology.  What I find really interesting is that not being able to detect this difference is the normal outcome.

[Ethan Winer]

Of course the rise/fall time and the intensity make  differences, too.

Yes, excellent point. If the volume or frequency response varies very slowly you can probably change it quite a bit before the change is noticed. The same goes for artifact audibility such as jitter, truncation distortion, and other forms of distortion. When I prepared the artifact audibility tests for my AES Audio Myths video, and again later for my upcoming book, I switched the noises on and off at regular intervals, to make them easier to hear, biasing the results in favor of those who believe such soft sounds are audible.

BTW, I asked in another forum for more / better research about the audibility of volume and frequency response changes, and someone posted this list of past research:

Code: [Select]

Study Authors     Year Published     Min. Detectable Fluctuation
Reisz                  1928             ~1 dB
Dimmick & Olson        1941             JND = 1.5 dB to 3 dB
Atal, et. al.          1962             ~ 1 dB
Jestaedt, et. al.      1977             JND @ 80 dB = 0.5 dB
                                        JND @ 5 dB = 1.5 dB
Toole and Olive        1988            .25 dB for a 5kHz resonance, Q = 1
References:
B. Atal, M. Schroeder, K. Kuttruff, "Perception of Coloration in Filtered Gaussian Noise Short-time Spectral Analysis of the Ear", 4th ICA , Copenhagen , Denmark 1962, paper H31.
F.L. Dimmick and R. M. Olson, '`The Intensive Difference Limen in Audition" JASA, vol. 12, pp. 517-525 (1941)
W. Jesteadt, C. C. Weir and D. M. Green, "Intensity Discrimination as a Function of Frequency and Sensation Level" JASA, vol. 61, pp. 169-177 (1977)
R. Reisz, "Differential Intensity Sensitivity of the Ear for Pure Tones", Physical Review, vol 31, pp 867-875 (1928)
F.E. Toole and S. Olive, "The Modification of Timbre by Resonances: Perception and Measurements", JAES vol 36, # 3, March 1988, pp 122-142
R. Hellman, et. al., "Determination of Equal-loudness Relations at High Frequencies", Department of Psychology and Institute for Hearing, Speech, and Language, Northeastern University, Boston, MA USA[/quote]

[stereocilia]

Good stuff.  I remember hearing a lecturer in psychoacoustics describe how the precedence effect could be used to fool a listener into perceiving sound at the wrong loudspeaker of a pair.  I don't recall the details, but it involved gradually shifting the signal from one side to the other without the listener noticing.  So, I guess the JND could be unlimited if we are allowed to play some psychoacoustic tricks.

[BPoletti]

I already provided you with a link providing evidence that differences as small as 3 millibels can be audible.  Repeatable results in testing indicate that the test was valid. 

What you refer to as "conventional wisdom" is simply not.  Period.

Have you asked John Atkinson to comment?  I'm sure you would get some interesting responses from the Stereophile forum. 

[Ethan Winer]

I already provided you with a link providing evidence that differences as small as 3 millibels can be audible.

I saw no evidence, only an anecdotal claim with nothing to back it up. As I said in the capacitors burn-in thread, "I see nothing about test methods, frequencies used for the source, number of people tested, how many times, blind or not blind, and so forth. Just one guy making a statement in an audiophile magazine. Versus the links I posted, two of which are on university web sites." I since posted more real research results in this thread. Further, I've tested this myself many times. Again, I'm glad to change my opinion if presented with real proof, documented proof, showing hard data including how the tests were conducted. If you have that, I'd love to see it.

--Ethan

[stereocilia]

Good stuff.  I remember hearing a lecturer in psychoacoustics describe how the precedence effect could be used to fool a listener into perceiving sound at the wrong loudspeaker of a pair.  I don't recall the details, but it involved gradually shifting the signal from one side to the other without the listener noticing.  So, I guess the JND could be unlimited if we are allowed to play some psychoacoustic tricks.

Okay, it looks like I'm referring to the Franssen effect.  I knew I'd find it.  http://en.wikipedia.org/wiki/Franssen_effect

[Steve]

I don't want to take the Capacitors burn-in thread off topic, so I figured it's best to start a new thread. This is a great topic for The Lab because it addresses what types of changes can be heard, and what types are too small to worry about. Separating fact from fancy makes us more knowledgeable, and in turn smarter consumers.

The conventional wisdom is that 1 dB is about the smallest change in level or frequency response that people can reliably identify. This is also known in science as the Just Noticeable Difference, or JND. However, the JND for level and response changes depends on the frequencies present, as well as their absolute volume level. At loud volumes the JND is smaller than at very soft levels. Likewise, JND is smaller at frequencies where our hearing is most sensitive. In other words, we can identify smaller level changes when the music is loud enough to hear clearly. So under ideal conditions - when listening at a decent level in a room with no damaging reflections, or when using earphones - it's possible to identify level and response differences smaller than 1 dB.

The book, Introduction to the psychology of hearing by Brian C. J. Moore states that the smallest detectable volume change for broadband noise is between 0.5 and 1.0 dB. Section 19.2.1 of Sound Reproduction: The Acoustics and Psychoacoustics of Loudspeakers and Rooms by Floyd Toole shows that changes in frequency response as small as 1/3 dB can be heard under ideal circumstances. Both of these are scholarly works that refer to real research, versus the 0.03 dB JND claimed anecdotally with no further explanation by a hi-fi magazine writer linked in the other thread.

However, when discussing changes in frequency response, versus volume level, two additional factors come into play. First, and perhaps too obvious, the frequencies being boosted or cut must be present in source. You could boost 50 Hz by 20 dB on a recorded flute solo, but that won't be noticed because flutes contain no content at such a low frequency. If such a boost is heard, it's surely due to other content in the recording, such as air conditioner rumble or other noises unrelated to the flute itself.

Another factor is the bandwidth of the boost or cut. When dealing with broadband content such as a full music mix or pink noise, the JND for a boost or cut with a wide bandwidth (low Q) is smaller than when only a narrow range of frequencies is changed (high Q). This too makes sense, because more total energy is affected with a wide bandwidth. In Toole's book linked above, his JND of 1/3 dB is for a boost having Q of 1. When the Q is raised to 10 the JND rises to 3 dB before the response change can be noticed by listeners.

I will keep this paper as simple as possible for ease of understanding.
I appreciate your listing some sources, but those sources either do not address frequency response changes or in an ineffective manner. Let's also examine to see if changes in broadband spl level changes are linked to freqency response changes.

Brian C. J. Moore used broadband noise for his tests. However, noise vs instruments/music are quite different conditions, thus not applicable.
For instance noise does not have a harmonic weighting factor that musical instruments have. An example of a few harmonics weighting factor is given below.

(RCA Radiotron Designers Handbook, 1960, 26 engineers, page 608, from Olson)

Harmonic         Weighting factor
2nd                  1
5th                  2.5
9th                                                    4.5
14th                                                  7

This means any change in instrument's fundamental and its harmonics will change the characteristics of the instrument
while broadband noise is not affected.
Instruments are also easily recognizable, while broadband noise is not.
 
Another problem is that if one scopes noise, it is not steady state but varies slightly across the broadband, kinda like
a flickering flame. So low level noise testing is inaccurate under low level changes. 

The next source is Toole. Toole provides good, useful information. However when it comes to low level frequency response changes, I see Toole only mentions two variables to address, that of
sight and manufacturer. Since no other variables are addressed any low level results will be skewed. 

Lastly, Q/bandwidth, and if the instrument is in the lower bands. It is true that different instruments will be affected differently because of their fundamental and harmonic content. However, other forces
come into play for all instruments.

First, let's address Q vs the real world example of an RC network. It is true that bandwidth plays a role in perception. The wider the bandwidth, the less deviance to be perceived. So a Q of 20 has a bandwidth of approximately .07 octave, while a Q of 1 has a bandwidth of approximately 1.35 octaves.
(See http://www.sengpielaudio.com/calculator-bandwidth.htm )

However, an RC network has a bandwidth of many octaves. So any deviance is perceived at a much lower frequency response change than using even a low Q resonance circuit. Another deviance to be considered is the overall rise and fall time slopes that have changed.

A further consideration is masking, which the bass affects. Altering the bass affects inner detail retrieval, and affects Depth, soundstaging etc. So bass level is very critical.

In conclusion, it can be seen that the sources either:
1) broadband spl changes are quite different from frequency response changes.
2) testing conditions such as broadband noise are quite different from actual instruments/music
3) do not address frequency response changes at all
4) test examples are inappropriate (Q vs RC network)

Cheers.

[Ethan Winer]

^^^ Interesting. I wonder how slowly they have to pan from left to right for this effect to work.

--Ethan

[Ethan Winer]

Toole provides good, useful information. However when it comes to low level frequency response changes, I see Toole only mentions two variables to address, that of sight and manufacturer. Since no other variables are addressed any low level results will be skewed.

Good comments Steve. Actually, Toole's test reports in his later book Sound Reproduction addresses frequency response changes when listening to music, not noise. The music is described in his text as "close-miked, low-reverberation, pop/jazz recordings." So his JND difference for a response change with a Q of 1 versus 10 is directly applicable to this discussion.

an RC network has a bandwidth of many octaves. So any deviance is perceived at a much lower frequency response change than using even a low Q resonance circuit.

Agreed, though in the other thread you argued the opposite - that frequency response changes can be heard with a smaller JND than level changes. At some point the bandwidth becomes the full range of audible frequencies, and then it's the same as a simple level difference. Logically, it makes no sense to me that JND is smaller for frequency response than for level. If anything it would be the other way around, since level affects all frequencies, not just some.

For me, the real issue is that this has been researched many times over many years. In all cases the researchers found that changes substantially less than 1 dB are not noticeable. So clearly the burden of proof falls on those claiming otherwise. Yet again, I'll be glad to change my opinion if shown credible research that includes details on the test methods.

--Ethan