1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.

    Dismiss Notice

Audibility of distortion

Discussion in 'Sound Science' started by wakibaki, Feb 17, 2014.
  1. wakibaki
    I've just been reading some stuff on Roger Sanders site. http://sanderssoundsystems.com/technical-white-papers/esl-amp-bias-wp. Sanders Sound Systems make ESL systems, everything but sources.

    What particularly caught my attention was this... 'Keep in mind that for most humans, the threshold of distortion detection is around 3%. If very special test tones are used, some people can hear about 1% distortion. No test has ever shown that a human can hear distortion levels below 1% under any circumstances'.

    Being interested in the subject, I always read any material on audio perception that comes my way, but this is the first definitive statement I have ever come across regarding the audibility of distortion. I take it he intends harmonic distortion in this case.

    Does anybody have any information to support or contradict these statements?

  2. EthanWiner

    Yes. Distortion at 1% means the artifacts are only 40 dB below the source. Most people can hear artifacts 60 dB down with specific frequency combinations. I've tested 100 Hz and 3 KHz sine waves, and even my 65 year old ears can (just barely) hear the 3 KHz tone when it pulses on and off. This is explained in more detail in my AES Audio Myths video, in the sections starting at 24:00.

    Now, audio circuits don't generally produce 3 KHz distortion from a 100 Hz tone, but your quote said " below 1% under any circumstances."

    ev13wt and gjfs like this.
  3. wakibaki
    Thanks Ethan.

    So you would suggest that 0.1% is a more realistic detection level for the possible detection of any flaw? And presumably you'd have to go a bit below that to ensure inaudibility?

    3k is close to the maximum human sensitivity, 25~30dB better than the sensitivity @100Hz. I guess features like that make it difficult (although not impossible) to be proscriptive about audibility.

    If, however, I were to suggest that no flaw more than 80dB down (0.01%) would be audible, how would you react to that?


    Shame about Sanders error, I guess in his enthusiasm to discount some audio mythology he overstepped the mark a bit.
  4. bigshot
    I'm not as techie as you, Ethan. But when we are talking about harmonic distortion, the noise is following the signal, right? I would guess is the fundamental gets loud, the harmonic distortion gets loud and quieter goes quieter. So there would be two factors making it more difficult to hear THD than a steady sine wave... 1) the masking effect of the loud signal drowning out the low level harmonic distortion; and 2) the fact that the noise is going up and down in volume and pitch, not maintaining a steady and consistent tone. That's why 1% THD is usually cited as the threshold. Am I correct in this?
    Of course speakers and turntables produce more than 1% distortion usually, and they certainly can sound very good.
  5. esldude
    Yeah, my understanding from multiple sources is .1% for most forms of distortion with test signals and between 2-4 khz.  So getting below -60 db should be transparent at least for basic distortion.
    Other tests with a few spaced sine tones is that the level of minimum audibility rises  to .5% if I am remembering right.  It definitely goes up with extra tones. 
    Simple music rises somewhat higher.  More complex music rises higher still.  3-4% being quote for something like orchestral works.  Some have pegged it a bit higher still. 
    Here is an interesting old article where Bob Carver, Julian Hirsch and others from Stereo Review test out the parameters.  I think some of their numbers are too high.  But they follow the trend of hearing lower distortion with simpler sounds vs more complex ones.
  6. ab initio
    A quick glance at the article didn't seem to indicate that the discussion was strictly related to harmonic distortion. I imagine the sum and difference frequencies generated by intermodulation distortion could sound a little more prominent that the harmonic distortion that tends to be masked. Maybe I can try making some test tones this week if I have some  time and folks are interested.
  7. EthanWiner

    Yes. Not only that, our hearing acuity improves at louder levels too. Further, if the noise floor in your room is 30 dB SPL, that 1% or 0.1% must be louder than the noise floor in order to hear it.

    A bass solo with distortion is mostly just a few sine waves. This is why I usually refer to "artifacts" rather than distortion.

    Yes, though masking is reduced when the frequencies are farther apart.

    It depends entirely on the source. Again, distortion on a bass solo might be more noticeable than a complex mix. Further, whenever you have THD you also have IMD, and IMD is more noticeable. Earlier I mentioned that circuits don't generally create a strong 30th harmonic and no other content. But IMD can easily occur at distant frequencies. For example, two flutes (which create fairly pure tones) playing a high D at 1,175 Hz and the E above at 1,318 Hz, when distorted, generate sum and difference frequencies at 2,493 Hz and 143 Hz. The lower frequency will certainly be noticed because it's not masked by the source tones at all.

    gjfs and whitedragem like this.
  8. Seifer01
    I'm not sure if this is the right thread for this post, but my recently purchased beyerdynamic dt990 pro's seem to be revealing quite a bit of distortion on my CD's. It's very noticeable with vocals, a sort of bubbling, rattling sound. is that distortion?
    My Senn hd429's reveal the same thing in the same places on the discs but to a less noticeable extent.
    It's pretty distracting because it adds a coarseness and grittiness to music.
  9. Speedskater
    Doc Earl Geddes has several papers Sound Quality and Distortion Perception:
    We also have to realize that "THD" attempts to put a simple number value on a very complex event.
    gjfs likes this.
  10. gjfs
    Totally false. Make a signal that's a mix of 40 Hz and 3 kHz, where the 3 kHz tone is -80 dB lower than the 40 Hz tone. This is equivalent to 0.01% THD (at the 75th harmonic!), and you'll be able to hear it pretty easily. You don't even need IMD. (I've tried to attach example tones to this post.)

    Is this a realistic scenario? Yes. I've seen power amps with crossover distortion that produce high frequency harmonics like this, and it was audible, despite not showing anything out of the ordinary on a numerical THD+N measurement. Worst case was the decay of piano notes, since the slow change in level caused a slow change in which harmonics were emphasized, and human ears pick up on "filter sweep" sounds easily. https://books.google.com/books?id=XOvf30iChsYC&lpg=SA2-PA129&ots=DdsIAR80Xe&dq=decaying piano&pg=SA2-PA129#v=onepage&q=decaying piano&f=false
    Last edited: Nov 7, 2017
  11. bigshot
    Your example of 40Hz and 3kHz tones is designed to make it as audible as possible. It's forcing the conclusion. Distortion with tones is different than distortion with music. You would never run into anything remotely like that in real world distortion in recorded music. In practice the 1% rule of thumb is pretty accurate. In one of the videos in my sig, Ethan Winer takes a horrible buzzing noise and lays it in under a music track. He drops it in volume level by level. It disappears long before -80dB. In the real world, distortion levels in digital audio are so low and so inconspicuous, they aren't even worth talking about. Inaudible is inaudible.
  12. gjfs
    Yes, but the scenario is based on real-world experience with customer complaints about piano tones through power amps that measured at 0.1% or better. The link in my previous post says the same thing.
    Last edited: Nov 7, 2017
  13. bigshot
    How do you know it has to do with the amp? A piano with percussive pure notes is hard for transducers to reproduce without distortion too. I'm not sure that lowering the distortion of the amp another digit to the right would help any.
  14. amirm
    If you mean this test (concerto): http://ethanwiner.com/audibility.html, I just took it and I can easily pass double blind test down to -60 db or 0.1%. I could probably pass the -70 db (his lowest level) also but need the original file to do it fairly.

    Because Ethan's test uses pulses, it is more revealing than it would be otherwise. And he provides for training at elevated levels which makes it even easier.

    Are you not able to hear it at that level? 1% is his -40 db and Ethan's version of that is so audible that it jumps out and grabs my throat. :)

    As to the general topic, single number figures of merit are useless because they are psychoacoustically blind. We can only determine audibility if we have the full spectrum of distortion and we then look at the threshold of hearing, masking, etc. and determine likelihood of the distortion being a problem. Earl's paper that was posted earlier talks some about this although he then goes down the hole of defining yet another single number figure of merit.
  15. bigshot

Share This Page