[Joe Bloggs]

Don't think so?  That's a graphic EQ, xnor mentioned using a parametric EQ.

 

[xnor]

I used the built-in parametric EQ in Adobe Audition, if that was the question, but any parametric EQ with a min. phase mode (the default) should work equally well if not even better.

 

[slwiser]

Quote:

Don't think so?  That's a graphic EQ, xnor mentioned using a parametric EQ.

See this thread has been confusing...waiting better info...thanks

 

[LFF]

Quote:

See this thread has been confusing...waiting better info...thanks

A graphic equalizer is the simplest type of EQ. A graphic eq consists of multiple sliders or controls for boosting or cutting frequencies of sound that have been predetermined by the manufacturer. A parametric equalizer is more complex and can control three aspects of each frequency: level (boost or cut), the center or primary frequency and the bandwidth or range of each frequency. Most also have built in low and high shelf controls. The one recommended for FOOBAR is a graphic. The one xnor used to change the CSD is a Parametric.

 

[dwk]

Quote:

 
I've spoken with Matt Ashland at JRiver about coming up with usable IR files to put into the MC17 convolution filter. I've had limited success so far. I need much more time to dedicate to this project to make it work, and being the curious type, I want to make sure it works with my speakers in the first place. In theory it should work, but there are quite a few issues:
 

• The IR files need to be massaged into what we as humans hear instead of what the microphone/measurement system ears. While my measurements make an attempt to do this, they are only relative approximations based on comparisons with other reference playback systems.
• I would have to have to compensate accordingly IR --(FFTs)--> FR --(EQ)--> FReq --(iFFT)--> IR. (Or the raw IR could be used in the convolution filter and EQ/compensation applied during playback)
• Even then I would have concerns about the IR being obtained from steady state measurements, increased distortion (headphone response can be really wonky), and obliterating the designers intended voicing of the headphone.
• I need to write some code to convert IR in delimited text to a WAV file since that's what MC17 takes.
• A driver that wants to resonant like a drum is going to do so no matter what, despite DSP
• DSP convolution with an IR file may be a bit too much overkill, possibly causing other issues, because of the time scale and specific headphone sensitivities involved. DSP room correction works on the scale of 10s of milliseconds. DSP headphone correction would have to work on the scale of fractions of a millisecond. My general philosophy is to use the simplest tool to get the job done in an acceptable way. PEQ may be much simpler.

 
So far, this project is on hold. But it is interesting enough for me to pursue in the very near future.

 
Very cool. Inspired by folks reporting that the Smyth Realizer actually really does work well, I picked up a set of in-ear binaural mics a while back with the intent of doing very much what you describe above. Additionally I wanted to experiment with HRTFs and Ambiophonics-over-headphones which I think could be very interesting.  I haven't had any time whatsoever to start down the road though.
 
 I have had a fair bit of success with the DRC software (drc-fir.sourceforge.net) for speaker/room correction. You're correct that the two problems are rather different, but since DRC is parameter driven it may well be possible to hack together a workable config for phones.  I was going to try starting there for lack of any other sophisticated filter generation package that I'm aware of.
 
 I think the main question in my mind is how to figure out what the 'ideal' FR is for a headphone. My assumption is that it is not flat, and should compensate for the difference in HRTF between something firing straight into your ear from the side and something firing from ~30 degrees off center out front which is where most recordings will be eq'd for.  I *think* the Smyth realizer process gets to more or less ignore this though, since it's measuring the effective HRTF in both cases, so they automatically cancel out of the problem.  It seems viable to do something along the lines of measuring a speaker response from both positions to deduce the FR difference - it would be a bit more cumbersome and would probably involve correcting the speaker to begin with, but if my thought experiment is correct is should result in something resembling the desired target response..
 
 Anyway, really interesting problem that I'm hoping to spend some time on, but probably won't have much free time until the fall.
 
On a different topic, I was planning to pick up some closed cans for a system at work, and was thinking of the FA-003 or the Brainwavez equivalents. Based on your CSD results I'm reconsidering going down the T50RP route. I didn't really want to get sucked into modding, but the results look fairly promising.

 

[arnaud]

Very cool. Inspired by folks reporting that the Smyth Realizer actually really does work well, I picked up a set of in-ear binaural mics a while back with the intent of doing very much what you describe above. Additionally I wanted to experiment with HRTFs and Ambiophonics-over-headphones which I think could be very interesting.  I haven't had any time whatsoever to start down the road though.

 I have had a fair bit of success with the DRC software (drc-fir.sourceforge.net) for speaker/room correction. You're correct that the two problems are rather different, but since DRC is parameter driven it may well be possible to hack together a workable config for phones.  I was going to try starting there for lack of any other sophisticated filter generation package that I'm aware of.

 I think the main question in my mind is how to figure out what the 'ideal' FR is for a headphone. My assumption is that it is not flat, and should compensate for the difference in HRTF between something firing straight into your ear from the side and something firing from ~30 degrees off center out front which is where most recordings will be eq'd for.  I *think* the Smyth realizer process gets to more or less ignore this though, since it's measuring the effective HRTF in both cases, so they automatically cancel out of the problem.  It seems viable to do something along the lines of measuring a speaker response from both positions to deduce the FR difference - it would be a bit more cumbersome and would probably involve correcting the speaker to begin with, but if my thought experiment is correct is should result in something resembling the desired target response..

 Anyway, really interesting problem that I'm hoping to spend some time on, but probably won't have much free time until the fall.

On a different topic, I was planning to pick up some closed cans for a system at work, and was thinking of the FA-003 or the Brainwavez equivalents. Based on your CSD results I'm reconsidering going down the T50RP route. I didn't really want to get sucked into modding, but the results look fairly promising.

I have also been contemplating doing this myself... For the ideal response, it is whatever you want it to be. Ideally, it should be either an HRTF recorded in a reverberant chamber (diffuse field personalized eq.) or anechoic chamber at specific speaker heading (free field personalized eq.). I would not say the realizer does anything wrong but normally you characterize the head response independent of the room or speaker characteristic (the HRTF is the pressure level at the ear canal entrance normalized by the spl measured without head/torso at that location).

Since you're going to use the hrtf for headphone playing, the headphone response (just like the speaker used to measure the hrtf) must be equlized out). I know the devil is in the details, like trying to make a stable / causal impulse response out of a headphone response with severe notches (null pressure at the mic location for that headphone/placement/head). The inverse of that is a peak with high Q (low damping), which needs infinite time to decay in the time domain (think a looong ridge in the csd). Try to convert that equalization curve back to the time domain, you're in for for funny sounding equalization.

There is lots of research on this, and I think it boils down to smoothing the data (equalized target) such that it is a stable and meaningful filter. For instance, some features of the measured data are very peculiar like purrin said: move the phone a tiny bit, re-sit the mic and the HF response totally changes, thus invalidating the sharp filter you came up with...

 

[ZeNmAc]

Quote:

 
I wouldn't be so sure. Someone with a measuring rig and an appropriate equalizer might chime in on this and give a definitive answer.
 
CSD plots are a function of the impulse response, which in turn are a function of the FR magnitude and phase. The way I see it, certain types of equalizers (which can tackle FR magnitude and phase) might be able to "fix" a bad CSD to some extent. Obviously there are limitations to what an equalizer can do, and real world proof would have the ultimate say... and currently I don't have a measuring rig... So you can label this as arm chair theorizing if you choose.
 
That said, the new Beyers seem to have way too many issues. I would say best to get better headphones that to try equalize an expensive and problematic headphone. Talking from experience on this one: Keeping my KSC-75, sold my DT-990 250 ohm (a bit too bright and were not getting enough attention from me). Furthermore, the DT-990 250 ohm is considered less offensive (sonically and measurement wise) than the T5P by people I trust, therefore I can tell you I have no intentions on buying or even trying the T5P (among other Beyer offerings). 

 
Quote:

EQ cannot fix/change the physical/mechanical limitations of a driver.  Using a turbocharger on a 4 cylinder to make 400hp doesn't make it a 400hp V8.  EQ/DSP only takes care of half the problem.  
 
Ask any Apple lover.  Homeopathy=/=Chemotherapy.  

 
This.  Decay comes from the physics of the setup.  You cant fix a mechanical problem with an electrical solution.
 
Quote:

Did a quick test:
 

 

 
Second plot shows the same headphones' CSD, but with a simple parametric EQ (one filter with a cut at 4300 Hz) applied to the signal used for the impulse response measurement.

Look at 4300hz.  Notice the decay isn't any faster.  The difference is the magnitude is smaller at time = 0s.  From there the decay is at the same rate as before.  Like I said, this is a mechanical problem caused by the driver/enclosure.  No amount of dsp can fix that.
 
That's why you can take headphones that have bad ringing and eq the heck out of them and come up with some crazy eq to get rid of the nasty peaks you hear, but that still won't make them sound any clearer.  With slow music you'll still have a big dip in the freq. response, and fast music still won't sound clearer.

 

[SanjiWatsuki]

Quote:

 
 
This.  Decay comes from the physics of the setup.  You cant fix a mechanical problem with an electrical solution.
 
Look at 4300hz.  Notice the decay isn't any faster.  The difference is the magnitude is smaller at time = 0s.  From there the decay is at the same rate as before.  Like I said, this is a mechanical problem caused by the driver/enclosure.  No amount of dsp can fix that.
 
That's why you can take headphones that have bad ringing and eq the heck out of them and come up with some crazy eq to get rid of the nasty peaks you hear, but that still won't make them sound any clearer.  With slow music you'll still have a big dip in the freq. response, and fast music still won't sound clearer.

It seems like you can hide the flaw, though. It's much more difficult to perceive a dip, compared to a peak.  It seems as long as you keep the correction within reasonable terms, it should mostly, although not solve, hide a mechanical issue while not having a great change to your sound quality because of how your brain fills in the gaps. 

 

[arnaud]

That's the puzzling bit, the original phone rings at 4kHz, the compensating filter rings just as bad at the same frequency, the two convolved together give you a clean decay at the frequency. If it looks clean like that in the CSD, the impulse response is bound to look cleaner as well, isn't it?

This is getting me itchy to revisit my CSD routine

 

[arnaud]

Actually, what really should be done here is looking at the reverberation time (time for the impulse to decay of say 60dB for various octave bands). This is actually the way I work in my field of sound and vibration prediction / control (and never look at CSDs actually). XNOR, if you're up for it, I wouldn't mind accessing your raw and filtered impulse responses...

 

[xnor]

Quote:

 
This.  Decay comes from the physics of the setup.  You cant fix a mechanical problem with an electrical solution.
 
Look at 4300hz.  Notice the decay isn't any faster.  The difference is the magnitude is smaller at time = 0s.  From there the decay is at the same rate as before.  Like I said, this is a mechanical problem caused by the driver/enclosure.  No amount of dsp can fix that.
 
That's why you can take headphones that have bad ringing and eq the heck out of them and come up with some crazy eq to get rid of the nasty peaks you hear, but that still won't make them sound any clearer.  With slow music you'll still have a big dip in the freq. response, and fast music still won't sound clearer.

I'm sorry to disappoint but the decay is as fast as it can get. A 'artificially' created perfectly minimum phase FIR filter with the same magnitude response shows the same decay. So the FUD's inappropriate here.
 
Also, the slow/fast music sentence doesn't make any sense to me.

 

[Joe Bloggs]

Like I said, I would love to see you create a perfect minimum phase inverse filter for the magnitude response, convolve it with the impulse and see how close to ideal the resulting impulse and CSD gets.
 

 

[arnaud]

Yes, if truly headphones response and their inverse are minimum phase then it should create a near perfect impulse response shouldn't it. That's it, I got to fire up matlab! These operations are like 3 lines of codes, I am just such a lazy bum!

 

[xnor]

It's not perfect but that's not surprising considering I used the raw data:
 

 

[Joe Bloggs]

Quote:

 
It's so close to perfect as to be hardly recognizable as a waterfall chart...
Quote:

It's not perfect but that's not surprising considering I used the raw data:

 
What other kind of data would there be for you to use?