Which is the highest frequency that You can hear? Options

[bandpass]

I might miss something, but I don't think silence needs a special encoding

It probably doesn't; however, to introduce any unnecessary uncertainty into an experiment is foolhardy.

Sorry, but I thought that fading was a basic operation. I didn't know it would need special care (at least regarding artefacts)

Science always requires extreme care. How 'basic' the subject of a mistake is, has no relation to the significance of its consequences (see, for example, here).

Designing an experiment around perceiving the presence or otherwise of audio, and unnecessarily introducing artefacts such as these is again, foolhardy:

[C.R.Helmrich]

I'm with bandpass here. Sorry if I'm too repetitive and educative, but we're on HA.

Why don't you guys add white background noise to the test samples? I.e. let the listener compare low-level noise and low-level noise plus a tone with fade-in and -out? That way both a) the artifacts bandpass showed and b) aliasing tones due to bad resampling at the listener side could be made irrelevant. In other words, we could have saved ourselves pages of speculation whether certain 22-kHz-hearing poll participants had configured their PCs properly because it wouldn't matter.

And by the way, one more argument for background noise is that it would also mask any low-level aliasing {or non-linear distortion} introduced by the human auditory system or the transducers in case of ultrasound stimuli (if such aliasing occurs; don't know if it does, but with background noise we're on the safe side). I think that's why some Japanese scientists in an article described adding background noise in their tests for audibility of up-to-28-kHz tones. I'll try to dig out the paper. {Here it is.}

So, in conclusion, I suggest you go ahead with bandpass's approach, but also add low-level white noise (preferably the exact same noise) to every sample, including the silence.

Chris

P.S.: Yes, I mentioned the background noise issue before.

This post has been edited by C.R.Helmrich: Mar 3 2012, 11:16

[IgorC]

More than fair enough. As for me it's not a problem to re-open a poll one more time. But previously let's create the files and discuss possibe issues. Feel free to upload new files. I would do it right now but I won't be home anytime soon.

This post has been edited by IgorC: Mar 3 2012, 12:07

[bandpass]

How does this look?

CODE

for f in 0 $(seq 8 23); do
  sox -R -n -b16 ${f}kHz.flac synth sin ${f}k noise remix 1p-6,2p-46 fade h .25 5 .25
done

which:

• has 6dB of head-room
• puts masking white-noise 40dB below the test tones
• -R ensures that the noise is generated the same each time
• uses flac to save some space

[C.R.Helmrich]

Sounds good overall, but some remarks:

• I'd leave out the 23 kHz sample, since it's beyond 22050 Hz. I think it wasn't in the test anyway.
• Are the resulting files mono? If not: they should be so that we don't have stereo noise and mono tones.
• I think we should let the tones start and end a few msec into the file, in case a listener's player chops off some samples (which might lead to clicks again)
• To save space maybe it's better to use 8-bit WAV (without dither, the noise already dithers). FLAC is still not supported by every player.

Sorry, I tried to upload my own samples here, but it doesn't work at the moment.

Chris

[musicreo]

I can hear up to 18 KHz. But with ABX test I can identify 19-22KHz. foobar Wasapi- Asus U1- Sennheiser HD 555. What is wrong with the setup or samples?

[bandpass]

• I'd leave out the 23 kHz sample, since it's beyond 22050 Hz. I think it wasn't in the test anyway.
• Are the resulting files mono? If not: they should be so that we don't have stereo noise and mono tones.
• I think we should let the tones start and end a few msec into the file, in case a listener's player chops off some samples (which might lead to clicks again)
• To save space maybe it's better to use 8-bit WAV (without dither, the noise already dithers). FLAC is still not supported by every player.

Yes, they're mono.
Worried that 8bit might not have enough dynamic range, so sticking with 16-bit wav and addressing the other points gives:

CODE

for f in 0 $(seq 8 22); do
  sox -R -n -b16 ${f}kHz.wav synth sin ${f}k noise remix 1p-6,2p-41 fade h .25 5 .25 pad .25 .25
done

I also have problems uploading samples, so please go ahead anyone who doesn't!

[extrabigmehdi]

Designing an experiment around perceiving the presence or otherwise of audio, and unnecessarily introducing artefacts such as these is again, foolhardy:

Humm, I'm a bit surprised. According to the graph and colors, the artefacts are in the -80 db -100 db region... (blue-pink colors) .
So at least this shouldn't affect a "normal use" of the fade, i.e in music.
Have you tried to "analyse" the fade generated by sox, you don't see such artefacts ?

I'll redo the hearing test later...

[onkl]

Why don't you guys add white background noise to the test samples? I.e. let the listener compare low-level noise and low-level noise plus a tone with fade-in and -out? That way both a) the artifacts bandpass showed and b) aliasing tones due to bad resampling at the listener side could be made irrelevant. In other words, we could have saved ourselves pages of speculation whether certain 22-kHz-hearing poll participants had configured their PCs properly because it wouldn't matter.

I'm curious. Why would the noise prevent aliasing of high frequency tones?

[C.R.Helmrich]

It wouldn't prevent it. It would mask the aliasing.

Chris

[extrabigmehdi]

Seriously I didn't hear any "artefacts" in the fade. The fade in / fade out are relatively fast (250 ms) , you don't even have time to realize there's a fade in / fade out by the way.
I avoided aliasing by configuring soundcard output , or just using sox.
And you guys are over-complicating things.

[greynol]

I don't think my samples are going to cause any false positives, but If they can be replaced with better ones then that should be done. Regarding the idea that noise will mask SRC artifacts, it better be a lot of noise, since some implementations can be really bad (older creative cards come to mind). As such, I think 48k sampling rate should still be used and other caveats should still be included.

Anyway, feel free to upload new and improved samples to the upload section under a new topic and I'll handle the the rest. I will not be generating them myself.

[greynol]

Could you please approximately compare the "grossly high levels" to some sound in real life?

Sorry, I can't. We are talking about pure tones at the edge of human hearing. This is not typical of what you might encounter in everyday life, except for something like the tone coming off a CRT.

[C.R.Helmrich]

Wait, the poll says "Total Votes: 74 - You have already voted in this poll", where neither of this is true. I count only 56 votes.

Anyway, let's see how the last poll ended: total number of listeners: 122, listeners voting 18 kHz and above: 29 (26.6 %)

The results of the current poll so far are: total number of listeners: 56, listeners voting 18 kHz and above: 18 (32 %)

So we have been very successfull in eliminating hypothetical false positives I'll let the others decide if it's worth opening another poll...

Chris

This post has been edited by C.R.Helmrich: Mar 3 2012, 18:09

[greynol]

listeners voting 18 kHz and above: 18 (32 %)

So we have been very successfull in eliminating hypothetical false positives

...and I can arbitrarily cherry-pick data and be sarcastic about how we haven't eliminated hypothetical false positives.

[C.R.Helmrich]

Sorry, wasn't my intention to criticize your and Igor's work. If I create a group "19kHz and above" instead of "18kHz and above", that group percentage-wise hasn't decreased in the current poll either. So the previous poll seemed good enough - or at least the listeners educated enough. That's what I tried to get across.

Chris

[IgorC]

Anyway there is still sharp drop from 18 kHz (10 votes) to a single(!) vote for 19 kHz. And there are still more votes for 20 kHz than for 19 kHz as I've talked before
Considering some young guys (20-26 years old. I know them from previous public tests) who have reported <19 kHz as max. frequency it's hard to beleive that some members actually hear a pure tones at 20 kHz and above (plus those members are somewhat older. 30-40 years old). That's my personal conclusion with which somebody is free to be agee or not.

This post has been edited by IgorC: Mar 3 2012, 19:19

[C.R.Helmrich]

Anyway there is still sharp drop from 18 kHz (10 votes) to a single(!) vote for 19 kHz. And there are still more votes for 20 kHz than for 19 kHz as I've talked before

I can understand the strong drop from 18 to 19 kHz - drops above 16-17 kHz have been reported in the literature, and it might be smoothed a bit by more participants. But the "peak" at 20 kHz is strange indeed. But maybe that's simply coincidence. I'd guess the participants in both polls are (will be) roughly the same.

Congrats to 1000 posts

Chris

[IgorC]

Congrats to 1000 posts

Thank You. I haven't noticed that. What a waste of moment.

[greynol]

I think it's important that we do our best to eliminate doubt.

If the results don't change does that make the exercise a waste of time?

[IgorC]

Well, I would try one more time again with the proposal of Chris. I guess it's way to learn.

[onkl]

It wouldn't prevent it. It would mask the aliasing.

Hmm. Aliasing sounds like a lower frequency. So let's say 20khz turns into 18khz. How would noise do any good here? I can only see it masking anything by being much louder, so you can't hear any high freuquencies at all.

This post has been edited by onkl: Mar 3 2012, 20:01

[extrabigmehdi]

Wait, the poll says "Total Votes: 74 - You have already voted in this poll", where neither of this is true. I count only 56 votes.

I don't remind, but there might be a "null option" for people that only want to see results.

[C.R.Helmrich]

Hmm, I checked what happens during bad resampling from 48 to 44.1 kHz. See attachment. Tones from 18 to 22 kHz are from left/green to right/pink, respectively. The white arrow points to the strongest aliasing component, which belongs to the 20-kHz tone. This might explain why more people voted "20" than "19" or "21"?

But onkl and greynol, you're right, masking such strong aliasing requires very loud noise, with a level equivalent to dither noise when going to 4-bit WAV. Not very nice.

Chris

[bandpass]

Okay, I've uploaded some new samples taking into account the above comments. In fact, I opted for masking noise at -30dB (instead of the -40 I suggested earlier).

@onkl, aliasing effects, unless a playback resampler contains no filtering at all, should be sufficiently small to be masked by this level of white noise.

Cheers,
Rob