Audibility thresholds of amp and DAC measurements

[j_j]

I have a question about the thresholds throughout the frequency range, are them really a flat line or should them be some kind of curve. For example does a THD+N of 0.005% have the same importance at 1kHz or at 50Hz. Thanks.

It's more basic than that. You must know the error spectrum. .005% can mean many, many completely different things.

 

[Blumlein 88]

To expand upon J_J's reply, and hopefully not misinform.

THD+N could be no distortion with -86 db noise over the whole audible band. It could be .005% all 2nd harmonic or .005% all 5th harmonic. One is sure to be masked and the other may not be. It could be several harmonics at a lower level that add up to .005% along with the noise. So the error spectrum would include harmonics with a level for each, noise levels, and you probably even need to know phase between the harmonics and the fundamental.

 

[trl]

I think that THD is more important than THD+N, but only if noise is not audible with your most sensitive headphones or speakers (when volume is as higher as you can usually listen), then I care not about the noise if I can't hear it when music is paused.

Feel free to consider this a subjective opinion, but I prefer a DAC or an amplifier with a better THD (e.g.: THD: 0.005% -> -86dB) and a higher background noise (e.g.: THD+N: 0.05% -> -66dB) than a DAC with a higher THD (e.g.: THD: 0.01% -> -80dB) and a lower noise (e.g.: THD+N: 0.02% -> -74dB). A little bit more noise will affect only the most sensitive headphones and speakers and only at very high output levels, while a bit more of distortions will affect the entire audio spectrum.

 

[Krunok]

I have a question about the thresholds throughout the frequency range, are them really a flat line or should them be some kind of curve. For example does a THD+N of 0.005% have the same importance at 1kHz or at 50Hz. Thanks.

Typically we tolerate much more distortion in bass region (say 20Hz-200Hz) than in the upper regions. IIRC we least tolerate distortions in the 1-5kHz region.

 

[JanRSmit]

Typically we tolerate much more distortion in bass region (say 20Hz-200Hz) than in the upper regions. IIRC we least tolerate distortions in the 1-5kHz region.

Depends on your hearing and damages attained during your live. I for one can suffer from serious sensitivity reduction due to distortions in the low spectrum. This is so since a hearing damage occured during my service in the army, as An anti-aircraft gun commander.

 

[Krunok]

Depends on your hearing and damages attained during your live. I for one can suffer from serious sensitivity reduction due to distortions in the low spectrum. This is so since a hearing damage occured during my service in the army, as An anti-aircraft gun commander.

For sure it does, but my statement was based on a healthy person with no incidental hearing accidents.

 

[Miguel]

Thanks everybody for your answers, but if it depends on error spectrum, noise or harmonics or something else then that graph doesn't mean a thing, i always thought that below some number of THD+N the distortion will be inaudible, independently of the spectrum of it, i thought it was guaranteed that below X we couldn't ear the distortion. That's why i was asking if that X was a constant with frequency or if it depended on it. By the way i was also curious about the crosstalk graph, is the thresholds constant or does it change with frequency.

 

[QAMatt]

mitchco said:
Measurements are meaningless if we don't know what is audible and what is not... And yah, there is a bell curve that would apply to most of us

Agree. There is a tool I helped worked on a few years ago located here. The tool allows you to start with a reference wav file of your choice, and then it will introduce different types of distortion on-the-fly (crossover, bit depth, amplitude, clipping, TANH). You can set the intensity of each type of distortion, do ABX comparison on the wav, and then vote. After 10 votes or so, you can sign your result cryptographically and post it in a forum if you wish. And then others can take your posted result and see if it's valid or not. So it's not possible to overstate your skill

The aim here was to see if there really are people that could, for example, hear the difference between 20 and 16-bit material (I cannot). But if someone has great ears AND a great system, then maybe they can. And hopefully we could find these unicorns and study them more.

If anyone would like to try the tool, just download and unzip--there's no installer. Pick a high resolution wav file (try here to get started), and then see how you do at reduced bit depth as a starting point. At a minimum, it's educational to learn what different types of distortion sounds like.

 

[solderdude]

@QAMatt Another test that could possibly be easy to implement is bandwidth limiting in the audible range with a somewhat, but not to steep, slope with a cut-off point which can be set between say 8kHz and 30kHz or so in a few 'steps' perhaps.

 

[trl]

Even 8bit vs 16bit test is really hard to compare with most compressed songs today: https://www.audiocheck.net/blindtests_16vs8bit.php. I will definitely try the ABX tool provided by Matt.

 

[QAMatt]

@QAMatt Another test that could possibly be easy to implement is bandwidth limiting in the audible range with a somewhat, but not to steep, slope with a cut-off point which can be set between say 8kHz and 30kHz or so in a few 'steps' perhaps.

Yes, agree, the ability to warp the spectrum would be very useful too. I just posted a new thread in case anyone wants to try and report their results.

https://www.audiosciencereview.com/forum/index.php?threads/qa_abx-test-results.8210/#post-204202

 

[j_j]

Thanks everybody for your answers, but if it depends on error spectrum, noise or harmonics or something else then that graph doesn't mean a thing, i always thought that below some number of THD+N the distortion will be inaudible, independently of the spectrum of it, i thought it was guaranteed that below X we couldn't ear the distortion. That's why i was asking if that X was a constant with frequency or if it depended on it. By the way i was also curious about the crosstalk graph, is the thresholds constant or does it change with frequency.

If total system error is better than 110dB, you're most likely safe, assuming a sensible gain structure. Was that the actual question?

 

[Miguel]

Yes that's the actual question, but is it always 110dB or does it change with frequency. I know that we hear with different sensibilities at different frequencies, how about distortion, does the minimal threshold changes with frequency.

 

[JanRSmit]

For sure it does, but my statement was based on a healthy person with no incidental hearing accidents.

Question is how to determine "healthy person with no hearing accidents". Would be very rare to find an adult that has no hearing damage of some sort. Also the way one perceives music of noise variies. Just look at the differences in liking of music.

 

[Krunok]

Question is how to determine "healthy person with no hearing accidents". Would be very rare to find an adult that has no hearing damage of some sort. Also the way one perceives music of noise variies. Just look at the differences in liking of music.

State of your hearing can be measured, medicine has solved that and defined thresholds for normal hearing for your age so that is an objective thing. Comparing that to a subjective thing like liking of music seems off to me.

 

[j_j]

Yes that's the actual question, but is it always 110dB or does it change with frequency. I know that we hear with different sensibilities at different frequencies, how about distortion, does the minimal threshold changes with frequency.

The meaning of the sum of all distortion energy being 110dB below your maximum listening level is not ambiguous. That's what an SNR of 110dB means for a signal at maximum listening level.

 

[Miguel]

Yes, but can we ear it. Do we need 110dB across the audible spectrum.

 

[j_j]

Yes, but can we ear it. Do we need 110dB across the audible spectrum.

You asked what I was SURE was safe. That's my answer.

 

[JohnYang1997]

There is difference between being able to hear a difference and hear distortion. At -100 to -120db range we still may perceive slight difference in sound sometimes. But when it's below -40db, -60db even, it's clean. We can't hear distortion without reference.
The absolute safe is when the amplitude of residues after the transducer is below human hearing threshold. Meaning, if you extra the distortion and noise out and play back them alone we can't hear it. That's the completely safe.
You don't NEED distortion below -60db to enjoy music. In fact the recordings are heavily distorted already so from subjective perspective you should be good with 1% distortion.

 

[j_j]

There is difference between being able to hear a difference and hear distortion. At -100 to -120db range we still may perceive slight difference in sound sometimes. But when it's below -40db, -60db even, it's clean. We can't hear distortion without reference.
The absolute safe is when the amplitude of residues after the transducer is below human hearing threshold. Meaning, if you extra the distortion and noise out and play back them alone we can't hear it. That's the completely safe.
You don't NEED distortion below -60db to enjoy music. In fact the recordings are heavily distorted already so from subjective perspective you should be good with 1% distortion.

Wrong.
My probe signal is a 19khz sine wave.
My noise signal is a 1.5kHz sine wave.

If you play the 19kHz sine wave back at 90dB SPL (assuming your system can do that), at what level can you hear the 1.5kHz sine wave?

You think it's inaudible at 30dB SPL?

Really? No.

You can do the same thing, give or take, with a 40Hz sine wave and a 3kHz sine wave as the 'error' signal.

Once you stick strictly to SNR (ThD, whatever) and don't specify the actual error spectrum, you've really widened the requirements, and yes, then you need to simply stay below absolute threshold of hearing, give or take.