Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[richard-ec2]

I have a slight hearing problem in that I'm over-sensitive to the "presence" range in audio playback, most especially frequencies in a narrow band around 4.4 kHz. (Apparently this kind of problem often goes with tinnitus, which I have too.) I've experimented in Adobe Audition and found that a notch filter of -40dB at 4.4 kHz does a good job of killing the offending frequencies while leaving the rest of the music intact. I want to find some way of inserting a notch filter like this into my playback chain but in such a way that it can easily be turned on and off, depending on who's listening and whether a particular piece of music needs it or not.

I think the best solution would be a filter in the DSP section of JRiver. I've experimented with the pre-installed parametric equaliser but I don't feel it's powerful enough to do what I want  - it doesn't seem to be as effective as the one in Adobe Audition so I suspect it doesn't go as deep as -40dB (even though it accepts that setting) and I don't think it goes as narrow as the Audition one either, even though it appears to accept any Q-width setting.

Does anyone have a better idea? I believe it's possible to install plug-ins into JRiver although I've never tried and I'm wondering if this is the way to go.

[mark_h]

Measuring pink noise on a FFT analyzer and adding a -40dB filter at 4400hz with a -40 gain and changing Q I can most definitely see the parametric equalizer in DSP doing what is asked of it...  but I guess you cannot find a Q narrow enough?

Try stacking the same filter in Parametric Equalizer 1 and Parametric Equalizer 2 - I was able to see narrow frequency attenuation this way than just using one EQ.

If the MC DSP doesn't get you there, then as you note, MC will work with external VST and there are plenty of options there, many free, many not so free...

[richard-ec2]

Maybe I'm doing the in-built parametric equaliser an injustice then. Do you know what maximum Q figure (minimum notch width) it will accept? You can enter anything you like but if I enter a very high number like 500, I can't hear an effect. I'm not sure if that's because it's no longer working or because the notch is now so incredibly narrow that I can no longer detect it.

[mwillems]

PEQ can create arbitrarily narrow filters (meaning that it will accept an arbitrarily high Q), but setting Q to 500 (!) won't create a useful filter for you.  A -3dB notch filter with a Q of 500 has a bandwidth of (more or less) 2 thousandths of an octave.  So at 4KHz a -3dB filter would be about 16Hz wide at best.  Your -40Db filter would be wider, but still on that kind of tiny scale.  

Presumably you're trying to notch out a slightly wider area than just 4392 to 4408Hz?  The human ear can't ordinarily resolve changes in frequency ranges that small (which is probably why it didn't "work"). Studies show that people have a hard time perceiving changes with bandwidths below 1/6 octave, but some folks can go a bit lower.  Anything much smaller than an 1/8 octave is unlikely to be very audible in most circumstances.  I'd start with Q's around 10 and move up as needed, but make the Q no higher than than 20 (which gives about 1/14 octave with a -3dB filter) and see how that works for you.

Also, if for some reason you need to stack filters, you can stack them in the same PEQ module, just add two or three filters.  The two banks are just for convenience and being able to customize the order of signal path, you can have an arbitrary number of filters in each bank.

[richard-ec2]

Thank you, that's very helpful indeed. I think what I'm aiming for is something that approximates the Adobe Audition effect show below  - sorry about the large size but at least you can see everything! It gives three options for notch width - narrow (as shown), very narrow and super narrow. Apparently "narrow" is a second order filter and "super narrow" is a sixth order filter. Unfortunately it doesn't give a Q figure for these settings and since I've no experience of this I can't guess what the Q figure is. But as you can see, at -40dB and on the narrow setting, it does spread quite well either side of 4400Hz and that degree of spread does seem to work quite well for me.




So what sort of Q figure do you think the image above is showing and is the JRiver PEQ capable of delivering it? If so, is there any advantage to be gained from using a plug-in or might I just as well stick with the PEQ?

[mwillems]

My guess would be that the Q of that notch is probably larger than 50 and lower than a 100.  Paddle around in that range and see if you can get it to sound "the same". 

Better would be to take measurements if you have a calibrated mic around.

[richard-ec2]

I've had a quick try and I think maybe nearer 100. 50 seems a bit too muffled so I think the notch may be a bit wider than in Adobe Audition.

The stacking idea - this is new to me but just thinking it through, is the idea to keep the notch narrower, on the basis that two stacked notches at -20dB each and a Q of 100 will be narrower than one notch at -40dB and a Q of 100?

[mwillems]

The stacking idea - this is new to me but just thinking it through, is the idea to keep the notch narrower, on the basis that two stacked notches at -20dB each and a Q of 100 will be narrower than one notch at -40dB and a Q of 100?

That's the idea of Mark's post I think, but I'd need to model to see if that actually would work as I've never tried it.  Some filters just add linearly (which wouldn't help you), but others do not.

[richard-ec2]

I'm trying to think through a means of testing for all these things - other than using my very unreliable ears!

I could record a pink noise file and play it back through JRiver, using whatever DSP settings I wanted to test. Then I could take the analogue output from the sound card of my PC and loop it back through the analogue input of the sound card of my PC, and record it to a new sound file. Then I could compare the new file to the old file  to observe what difference the DSP had made.

But using what software? It would have to be something very accurate to show up a very narrow notch at any degree of accuracy.

I'm also wondering if there's a way of cutting out the soundcard from the loop and keeping everything in the digital domain.

[Frobozz]

You can try some VST plug-in EQs to see if they work better for you.

Here's a freeware VST filter that you can try: RS-MET EasyQ
EasyQ uses a minimum phase filter and includes a notch style filter option.  You can also experiment with other EQ plug-ins that use different styles of filters (like linear phase).  I don't know of any freeware parametric EQ plug-ins that do linear phase.  There are some commercial VST plug-in EQs that will allow you to select different filters (like DDMF).  Try the EaseQ plug-in and try the DDMF EQs (like their LP10 EQ) and see if they work for you.  EQs for specific things like you're after are just something you'll have to try to find out what works for you and what sounds best.

[6233638]

• Download Audacity and generate a file containing white noise.
  
• Download and install the Voxengo Span VST plug-in.
  
• Play white noise on loop, adjust the filter, observe the effects in Span.

The image attached to this post will show you some useful settings for Span, and demonstrates a -40dB 4400Hz Q100 (100Q?) notch filter via Media Center's Parametric EQ
I'd recommend setting the Span plug-in to "process only when viewed".

[mwillems]

You could also just use the pink noise calibration clips built into JRiver (pink noise is generally preferred to white noise for calibration because it has equal energy per octave, white noise has linearly equal energy which doesn't correspond to the way we perceive sound).  

62 that SPAN plugin is pretty neat and I'll probably keep it around for analysis if I can iron out some kinks; do you know any way to turn off the A (or maybe C) weighting it's doing?

Never mind I found it.  The "slope" dial in the settings is the weighting curve.  Set it to 3dB so that pink noise measures flat and use JRiver's audio claibration function under tools-->advanced tools.  You'll be good to go.  Calibrating with white noise (beyond simple stuff like this) will result in vary strange sounding EQ generally.

It looks (to me) like the Audition notch you're trying to emulate has a Q around 70 (the 100Q notch is significantly narrower).

Hat tip to 62 for the sweet in-line RTA.  All of the RTA's I've been using require a microphone, and I've been looking for a decent "pre-out" RTA on and off for a while.   Thanks.

[6233638]

Well I thought it was the other way around, with white noise being spectrally neutral (flat line, or a slight slope upwards) while pink noise is perceptually neutral?
https://en.wikipedia.org/wiki/Colors_of_noise#Technical_definitions
 
But yes, you can use the slope control to level it out if you're using pink noise, and perhaps that would be better for these purposes.

[blgentry]

White noise is equal energy per frequency.  Pink noise is equal energy per octave.  Most frequency response measurements are made using swept tones or chirps or MLS or something like that these days and these are "flat frequency profile" measurements (as I understand it).  However, historically, most RTA type measurements *are* made using pink noise as a source.

It seems that this is somewhat subjective (the general concept) and that most hardware RTAs are already set up to look for pink noise as being a "flat" source.  I think the big deal here is real time analysis versus swept analysis.  I think real time is a flawed way of measurement and pink noise is a compensation.  But that's off topic for sure.

Brian.

[mwillems]

Well I thought it was the other way around, with white noise being spectrally neutral (flat line, or a slight slope upwards) while pink noise is perceptually neutral?
https://en.wikipedia.org/wiki/Colors_of_noise#Technical_definitions
 
But yes, you can use the slope control to level it out if you're using pink noise, and perhaps that would be better for these purposes.

White noise is equal energy per frequency.  Pink noise is equal energy per octave.  Most frequency response measurements are made using swept tones or chirps or MLS or something like that these days and these are "flat frequency profile" measurements (as I understand it).  However, historically, most RTA type measurements *are* made using pink noise as a source.

It seems that this is somewhat subjective (the general concept) and that most hardware RTAs are already set up to look for pink noise as being a "flat" source.  I think the big deal here is real time analysis versus swept analysis.  I think real time is a flawed way of measurement and pink noise is a compensation.  But that's off topic for sure.

Brian.

Brian's answer is the answer.  Pink noise is equal energy per octave, white noise is equal energy per frequency.  Because of the way our perception of sound works (we perceive tone separation logarithmically), pink noise is a better calibration target because it sounds perceptually flatter than white noise (because each octave has the same amount of energy).  If you filter white noise so you were listening to the octave between 500Hz and 1KHz, and A/B it with filtered white noise between 200Hz and 400Hz, the first will sound "louder" even though they have the same absolute SPL (the effect will be more exaggerated as you get farther and farther apart, but the sensitivity of our hearing (i.e. equal loudness contours https://en.wikipedia.org/wiki/Equal-loudness_contour) comes into play at the fringes in a way that can confound the comparison).

Obviously logarithmic sine sweeps are better for serious measurement (which is what I recommend in my correction guide: http://yabb.jriver.com/interact/index.php?topic=87538.0), but if you need to use RTA (for calibration or level setting) pink noise is the "right" answer.   Bob Katz weighed in on this at length when JRiver was standing up the R128 and loudness algorithms as I recall, and provided feedback on the calibration tones available in JRiver.

And because pink noise is "perceptually" flat, almost all RTAs show pink noise as flat (like Brian mentioned).  But I agree that real-time analysis is best reserved for quick sanity checks or level setting.  It's not as accurate as sweeps or other impulse based analysis, but will do just fine for something like this.  

[6233638]

almost all RTAs show pink noise as flat

Good to know.
Setting the slope to about 2.9 seems to be truly flat, but since it's 3dB/octave I suspect that the slope should be set to 3.0
That gives it a slight upward slope, but that seems to match white noise with a slope of 0. (due to the frequency spacing being logarithmic rather than linear, according to the Wikipedia link I posted above)

[richard-ec2]

Many thanks indeed for the advice, everyone. Much appreciated.

I think I'll play around with Voxengo Span because I already have Voxengo CurveEQ for other purposes and the user interface is the same so hopefully it'll feel familiar from the outset.

It's a pity you can't control the DSP options using JRemote but that would be greedy! I have thought of a way of doing it from the iPad, though - I can control the server PC (and JRiver) using Splashtop or another remote desktop app. If anyone has a better idea, I'd be interested to hear it.

[richard-ec2]

In fact (sorry to post twice in succession) that JRiver pink noise tool is really useful all by itself. If I just tell it to play endless pink noise, then go into DSP and start adjusting the Q width on my (-40dB at 4400Hz) notch, it's quite easy to get a result that's pleasing to the ear. Without the notch, there's a part of the pink noise spectrum that stands out above the rest and is irritating to my hearing (at 4400 Hz, obviously) so all I have to do is play around with the settings until the irritation just disappears, and no more. I'll try adjusting it at different times of day but right now, the frequency and gain seem to be right and a Q width of somewhere between 70 and 100 seems right, probably nearer 70 - although there may be an element of suggestion there!

So it's all working very well and many thanks again for the guidance.