USB Whine FIltering (Yeti Curse)

Special Interest Groups Audio Processing
21

If it is USB in origin then it will be 1kHz. A 1 mS polling period is written into the standard. However it probably isn’t an particularly accurate 1 kHz as that would be subject to the accuracy of clock source for the buss master, and there probably is a fair bit of AM & FM modulation on any sample, so you can’t make your notches too narrow, but there might be some gain if you can tune to the particular source.

The way the buss works is every 1 mS the master wakes up and polls all of the slaves for any data that they have to send or receive. In the typical case where the USB sound device has the bus to itself this results in a ~125 uS burst of data every 1mS. (Or two such bursts in rapid succession if the device is both recording and playing).

There are probably a myriad ways that the noise can get coupled into the microphone preamp. The data transmission is not quite truly differential, there is a buss state called “End of Packet” where both of the lines are pulled low at the same time, so there will be some common mode radiation from the data lines. Also the power consumed by the driving the buss will cause the power rail to sag and the ground rail to rise.

22

If it is USB in origin then it will be 1kHz.

What he said.

Also the power consumed by the driving the buss will cause the power rail to sag and the ground rail to rise.

I try and avoid saggy rails. Not chic. At all.

ringing artifacts

Thanks for the reference. I noticed that it uses an infinite fall-time step function to illustrate the point. If you don’t apply step functions in favor of normal speaking voices, can I assume the effect is a lot less extreme?

I see the ringing is pretty high pitched. Where does ringing for the 16000 notch go?

The illustration also uses one notch. If you use multiple notches as we plan to, what’s the possibility of some ringing artifacts falling into cousin notches and vanishing forever? There’s no shortage of harmonically related notches. And even if they don’t fall in and vanish, if you produce enough of them, don’t they just turn into white-ish noise?

And down to earth for a second, what’s the voice distortion or damage difference between 5 and 16 notches? Any of it audible?

Is that difference worth the additional operating complexity over, say, 10 notches fixed and and execution that always looks like:

Effect > Mosquito-Killer > OK.

Koz

23

Mosquito-Killer3.ny

What’s the Q of a selected notch of 8 at 8000? I can’t tell where the code is going and I can’t duplicate the shape of the notch manually.

Koz

24

I think the ringing is the same pitch as the notch.

The illustration is 2 notches versus 8 notches

The additional notches won’t alter intelligibility.
Steve did write code which will notch up to 1/2 the sample-rate, (so 22 notches on 44100kHz, 11 on 22050Hz ) …
kHz-Killer.ny (474 Bytes)
The notches above 16kHz aren’t going to make an improvement, so just a waste of processing time.

25

I’ve been going some real-world testing with Mosquito-Killer3 (8).

With setting of 8 instead of 5, I was able to search multiple instances of mosquitoes and vanquish all of them, including the killer nasty one from 2014. This is the one that originally called for 20 iterations of suppression to get rid of all the tones. I can’t tell if 9000 and up are still there and not objectionable, or I just can’t hear them.

I dug out the clean, WAV copy I have of a Sarah Vowell AudioBook CD and compared before and after voice quality. I can see ‘after’ holes in Spectrum View, but I can’t hear them while rapidly switching between the two versions.

http://www.kozco.com/tech/audacity/clips/SarahVowellBnB-Plain.wav
http://www.kozco.com/tech/audacity/clips/SarahVowellBnB-Mosquito8.wav

So with the idea of this being a lifeboat filter and some very tiny tonal shifting may be acceptable, I would like to move the default up to 8, bolt the Nyquist frequency limit code in there (that was Rev3, right?) and ship it. Rev4?

Koz

26

Is this progressing far enough to move to the Nyquist board for new plugins?


Gale

27

Indeed it is. This shows the ringing produced by Mosquito-Killer3 (default settings) when applied to white noise that ends abruptly.
notch-ring.png

28

Is this progressing far enough to move to the Nyquist board for new plugins?

I guess???

I keep waiting for someone to be horrified over some mistake.

Isn’t that where I write Purpose, Use and Comments? Shouldn’t I write that first?

Koz

29

Sounds like a fair compromise between utility & speed.

Jo[e] Public also refers to the problem as …
“USB mic high pitched whine” / “USB input high-pitched whine”
Only some compare it to a “mosquito”.

30

Only some compare it to a “mosquito”.

I bet it’s safe I’m the only one that calls it the Yeti Curse.

Since the whine has no easy, natural name, I suspect a majority (or all) the time we will be pointing a poster to it. Nobody is going to find it on their own. I guess if I’m clever with the description, someone could find it that way.

Koz

31

Nothing stops you from putting all of the names in the description. That way a search engine will find it.

32

all of the names in the description

I’m guessing Blue Microphones wouldn’t be too happy about publishing “Yeti Curse” in the documents. I might leave that one out.

Koz

33

You might as well leave it out.

There’s no way you’re going to get on top of the search results as this also seems to be a popular term in the WOW/Runescape gaming community.

Only when you search for “get rid of the “yeti curse”” the second link is to this forum. But that’s something most users will never do, imho. Too many words and it needs the quotes around “yeti curse”.

34

We’ll stick with ‘USB Microphone Whine’ and variations.
Koz

35

I’m doing this in a publisher/layout program, so this is copy/paste.



Mosquito-Killer4

PURPOSE

Some affordable USB microphones and other USB audio devices can add a constant, irritating, high-pitch whine sound (electronic mosquitoes) to your recording.

http://kozco.com/tech/audacity/clips/USBMicrophoneWhineClip.mp3

The whine sound is very difficult to remove using normal tools, but we found the personality of the whine didn’t change that much computer to computer, so we designed a Nyquist effect to remove it.


INSTALL
Mosquito-Killer4.ny (363 Bytes)
On-line Manual
Follow the link for your operating system then follow the instructions to install Nyquist effects.
http://manual.audacityteam.org/man/faq_installation_and_plug_ins.html#plug-ins



USE

Select the work you want to repair, or all of the work by clicking just above MUTE (on the timeline left).

Effect > Mosquito-Killer4 (8 mosquitoes) > OK.

That’s it. We expect the whine to vanish in one pass with no adjustments and little or no audible sound damage.



COMMENTS AND NOTES

Mosquito-Killer4 is a multi-point Nyquist notch filter designed to search and destroy the most common, objectionable tones in the whine sound.

1000Hz, 2000Hz, 3000Hz 4000Hz, 5000Hz, 6000Hz, 7000Hz, 8000Hz.
That’s usually enough.

The tones are caused by USB electrical housekeeping leaking into the sound.

Please be clear we are surgically removing very tiny portions of show and you may hear some very minor performance changes. The changes are usually so gentle compared to the piercing USB whine that nobody cares.


If you understand Analyze > Plot Spectrum…

http://manual.audacityteam.org/man/plot_spectrum.html

…or timeline Spectrogram View…

http://manual.audacityteam.org/man/audio_track_dropdown_menu.html

…or just keen to experiment, you may find that your whine has fewer than eight significant mosquitoes in it and you can use the effect successfully with a lower number.

The number is adjustable 1 thru 16. As a general production rule, the fewer corrections, effects, enhancements and filters the better.

Best of all is to use a microphone or sound system that doesn’t whine.

If you have troubles, post a message to the Audacity Forum.

https://forum.audacityteam.org/ucp.php?mode=login

.

36

Can you post the link to “Mosquito-Killer4”? I can only find links to earlier versions in this topic.

Gale

37

I’ll straighten that out. The nomenclature didn’t keep up with the code.
Koz

38

I didn’t check with the OED, but other authorities assure me both mosquitoes and mosquitos are acceptable. The “e” version looks better. There’s still one sentence that didn’t come out quite right.

There may also be a code error. I’ve been testing with a non-standard version.

Koz

39

The man text above is good to go. That’s the good news.

The less than good news is I can’t correct the code. The corrections are based on Mosquito-Killer3.ny.

There are two changes:
;name is 4 instead of 3.
;name “Mosquito-Killer4…”

;control iter is changed from what I perceive to be default 5 to default 8.
;control iter “How many mosquitos to kill ?” int “1 - 16” 8 1 16

I did it multiple different ways including cp, vi and cat. No dice. I can’t tell if the Mac is trying to help me sub rosa, or I just don’t know what I’m doing.

Koz
Screen Shot 2017-05-11 at 13.19.30.png
Screen Shot 2017-05-11 at 13.19.05.png
Mosquito-Killer4.ny (363 Bytes)
Mosquito-Killer3.ny (363 Bytes)

40

You’ve got a mix of LF and CRLF line endings, but that shouldn’t be a problem.



Both of those should be OK.

Woah! What’s that about?
Try trashing your pluginregistry.cfg and pluginsettings.cfg files, then re-install Mosquito-Killer4.