Dirac: Initial Experience

[LBob]

Just finished my first Dirac project. I have to say the experience was very straightforward. I went through the manual several times before the first try, and the whole project took only about 20 minutes from start to finish. I had allowed a couple of hours so I was pleasantly surprised. Here are the results for "couch." These were done with a UMIK 1 90 degree orientation using a calibration file created via rcohens helpful utility.

 

 

[ATTACH=CONFIG]

14910[/ATTACH][ATTACH=CONFIG]

14911[/ATTACH]

 

I knew I had some bass problems, but I was surprised when I saw the graphs. Overall though I thought the frequency response was pretty good above 100hz. I have no idea how to interpret the impulse response. I was a little surprised that there was a delay only for 1 speaker, but again I have not idea how to interpret these results.

 

Using the Dirac processor was also straightforward. All three of the players I used (Audirvana + 2.02, Amarra 3.0, and Pure Music 2.?) recognized the Dirac Processor and played through it. I unchecked integer mode and direct mode in Audirvana 2.0, but otherwise no changes needed. The first listening I did was through Amarra 3.0. It recognized Dirac as the output automatically which was nice. I like the ability to toggle back and forth between Dirac and straight player output. However, that feature is potentially misleading. When I first toggled back and forth I expect to hear seismic shifts in the sound. That didn't happen. In fact I thought I had screwed up something so I went back and created another filter. Same result. The differences became more apparent with extended listening rather than quickly switching back and forth. So far what I hear is:

 

1. Much sweeter strings. Massed violins on my system especially in the higher registers always had a "sandy" quality about them. No more. I have no idea why this would occur, but it's pretty consistent. I had always thought that this was just the digital violins sounded. This alone is worth the price IMO.

 

2. Cellos with more body. I think this might be due to the dip in the upper bass that Dirac corrected?

 

3. Increased front to back depth. Layering of the orchestral sections is much more apparent.

 

4. Imaging is more stable. Minor head movements don't make the oboes sprint towards the double basses.

 

5. Darker background. Noise decrease?

 

One thing that surprised me was that I don't really hear major differences in the bass. I suspect that the dips are nulls and that they aren't really correctable. It's also possible that the dips and valleys are pretty low. One thing that bothers me is that the graph shows a peak around 25 hz and I don't think that my speakers go anywhere near that level at that frequency.

 

So far I haven't played with the target curve. That's mostly because I don't really have an idea of what to try. Will re-read the manual. Any suggestions will be gratefully accepted.

 

Part 2 will be to try some new projects. First a chair at 90 degrees, Second and third, a couch and chair at 0 degrees. Then I'll see if there are huge differences.

[hellblau]

You may want to experience target curves similar to the ones below, wich are applied on my equipment/room.

 

1) A slight slope - let's say 1dB/octave from 100 - 200 Hz. This will look similar to the Dirac default curve.

 

2) Same as point 1., with a 2 - 4 dB dip in the 3 kHz region (sometimes called "BBC dip"). Quoting Electro-acoustic models : "Around 3 kHz our hearing is less sensitive to diffuse fields. Recording microphones, though, are usually flat in frequency response even under diffuse field conditions. When such recordings are played back over loudspeakers, there is more energy in the 3 kHz region than we would have perceived if present at the recording venue and a degree of unnaturalness is introduced.

This applies primarily to recordings of large orchestral pieces in concert halls where the microphones are much closer to the instruments than any listener. At most listening positions in the hall the sound field has strong diffuse components."

 

 

[LBob]

Thanks Hellblau. Very helpful. It looks like your speakers already had a BBC dip judging by the graphs. I remember Rogers monitors very fondly. I'll try it.

[hellblau]

You're welcome LBob.

 

My speakers are B&W 804 Nautilus (1st series, year 2000), and yes, they seem to have a kinda "factory BBC dip", though a little too deep, resulting in loss of detail in the high-midrange (crossover cuts at 4 kHz), at least in my room. The top-end models, like the 801, had the "Nautilus shaped head" including the midrange (same loudspeaker as mine) in a spherical structure, thus probably improving the dispersion and then the linearity of the room response around the crossover frequency.

 

Another important parameter is the "Global DSP gain" (misleading term in this case, since "gain" can only be negative. See "ab" in the picture below). The default value is -8 dB, but you should try to set the highest value you can before clipping occours in "close to 0 dB recording", like pop-rock records. I set -5 dB, and probably I could go even higher. Anyway, everything is explained in the user guide.

[Boris75]

Another important parameter is the "Global DSP gain" (misleading term in this case, since "gain" can only be negative. See "ab" in the picture below). The default value is -8 dB, but you should try to set the highest value you can before clipping occours in "close to 0 dB recording", like pop-rock records. I set -5 dB, and probably I could go even higher. Anyway, everything is explained in the user guide.

 

Is the clipping indicator now available also on the Mac OS X version? For me the lack of it has been a show stopper for me after I tried the software. The -8dB attenuation is too much, but on the other hand I don't want to go too much into clipping, but I need some indication.

[flak]

 

I have no idea how to interpret the impulse response.

 

 

I'm pleased by your positive findings

I had posted about the interpretation of the impulse response graph but I don't link it for you 'cause the images are not there anymore so it is now partially meaningless... if interested I can post it again in this thread.

 

Ciao, Flavio

[flak]

Is the clipping indicator now available also on the Mac OS X version? For me the lack of it has been a show stopper for me after I tried the software. The -8dB attenuation is too much, but on the other hand I don't want to go too much into clipping, but I need some indication.

 

Hello Boris,

the clipping indicator is not available on OS X because the system volume control is applied after the Dirac Audio Processor.

So even if the level out from DAP would be very high, it may still not clip if you don't have the system volume control at max.

 

Ciao, Flavio

[ecwl]

I think the better violins and cellos are not due to bass dip correction but are due to removal of the bass peaks. The uneven bass peaks often are not heard directly unless the individual notes are played repeatedly with other notes. But the subtle bass peaks can overwhelm other frequencies, giving a less natural sound.

 

As for the better depth and imaging, some of this also is related to the low bass being more even which re-creates soundspace. Walk into a mall and close your eyes and you can hear you're in a large space. Take out your iPhone RTA/AudioTools and you'll be surprised the low bass info and that's probably giving you the auditory cues for space. But getting two speaker delays to be identical definitely matters too for stable imaging. That's why some of us use laser distance measuring tools to make sure the speakers are equidistant from our ears.

 

At least that's my understanding of how things work. But I don't know this stuff that well. Hopefully, more experts will give a better and maybe even more correct answer.

[LBob]

Flavio,

 

I would be very interested in the post on interpreting impulse response graphs. I tried to google it but I couldn't find anything at a non-engineer level.

[LBob]

You're welcome LBob.

 

My speakers are B&W 804 Nautilus (1st series, year 2000), and yes, they seem to have a kinda "factory BBC dip", though a little too deep, resulting in loss of detail in the high-midrange (crossover cuts at 4 kHz), at least in my room. The top-end models, like the 801, had the "Nautilus shaped head" including the midrange (same loudspeaker as mine) in a spherical structure, thus probably improving the dispersion and then the linearity of the room response around the crossover frequency.

 

Another important parameter is the "Global DSP gain" (misleading term in this case, since "gain" can only be negative. See "ab" in the picture below). The default value is -8 dB, but you should try to set the highest value you can before clipping occours in "close to 0 dB recording", like pop-rock records. I set -5 dB, and probably I could go even higher. Anyway, everything is explained in the user guide.

[ATTACH=CONFIG]14919[/ATTACH]

 

Made a "BBC" filter for grins. Sounds quite nice. I lived with some small B&W's for years and liked them a lot. Ditto with a pair of KEFS. Ramped up the gain to -3. It got louder not sure about sonic differences.

 

Looking at my uncorrected frequency response I think that one reason there weren't major differences is that I pretty much had a response of a small bass boost with a gradual roll off.

[rcohen]

LBob, a couple suggestions based on my experiments. YMMV, so see if this works well for you:

 

1) On the left side, avoid that sharp right-angle change in slope between the measured response and the target curve. The difference between the two curves is the correction filter. When the correction filters have steep slopes with high magnitude, that causes ringing. This is most audible with the bottom frequencies as blurring out fine details. Instead, try to smoothly transition from the measured response to the target curve. This may help with the impulse response.

 

2) On the right size, when the treble angles up like that, it can sound harsh. Try zooming in around the right side of the graph, dragging right-hand limit beyond the rolloff point of the measurements, and create a target curve where the treble tapers down very slightly to the limit, rather than shooting up. Find the shape that sounds detailed and natural, but without harshness or fatigue. This side of the curve can be very sensitive to small changes, and the sweet spot varies depending on speaker and room.

[sebastien696]

LBob, a couple suggestions based on my experiments. YMMV, so see if this works well for you:

 

1) On the left side, avoid that sharp right-angle change in slope between the measured response and the target curve. The difference between the two curves is the correction filter. When the correction filters have steep slopes with high magnitude, that causes ringing. This is most audible with the bottom frequencies as blurring out fine details. Instead, try to smoothly transition from the measured response to the target curve. This may help with the impulse response.

 

2) On the right size, when the treble angles up like that, it can sound harsh. Try zooming in around the right side of the graph, dragging right-hand limit beyond the rolloff point of the measurements, and create a target curve where the treble tapers down very slightly to the limit, rather than shooting up. Find the shape that sounds detailed and natural, but without harshness or fatigue. This side of the curve can be very sensitive to small changes, and the sweet spot varies depending on speaker and room.

 

Interesting. Are you talking about the left end of the target curve ? To me it seems to follow the measured response ?

[Boris75]

Hello Boris,

the clipping indicator is not available on OS X because the system volume control is applied after the Dirac Audio Processor.

So even if the level out from DAP would be very high, it may still not clip if you don't have the system volume control at max.

 

Ciao, Flavio

 

Flavio,

 

Many thanks for your response.

 

Most people have the system volume set to max beause OS X volume control is not hi-fi quality at all: a clipping indicator assuming that the system volume is set to max would be very helpful.

 

Perhaps the Dirac audio processor could even check the system volume so that the clipping volume indicator becomes active only when the system volume is set at max level.

 

Boris.

[sebastien696]

Flavio,

 

Many thanks for your response.

 

Most people have the system volume set to max beause OS X volume control is not hi-fi quality at all: a clipping indicator assuming that the system volume is set to max would be very helpful.

 

Perhaps the Dirac audio processor could even check the system volume so that the clipping volume indicator becomes active only when the system volume is set at max level.

 

Boris.

 

While I agree it would be nice to have this clip indicator on Mac, I'm using Audio Hijack (but some others might do it) to check clipping. Dirac -> sound flower , Soundflower hijacked. (it works with system audio as well). It might not be totally accurate, but I found out (in my room configuration) I need to leave -8dB (maybe - 7dB) on some recent recordings with heavy bass.

[sebastien696]

@Flavio

 

Hello,

 

Could you tell us how is designed the "Autotarget curve". Is that a Dirac House curve, or is it calculated averaging the speaker response in the room ?

 

Thanks

[sebastien696]

LBob, a couple suggestions based on my experiments. YMMV, so see if this works well for you:

 

1) On the left side, avoid that sharp right-angle change in slope between the measured response and the target curve. The difference between the two curves is the correction filter. When the correction filters have steep slopes with high magnitude, that causes ringing. This is most audible with the bottom frequencies as blurring out fine details. Instead, try to smoothly transition from the measured response to the target curve. This may help with the impulse response.

 

2) On the right size, when the treble angles up like that, it can sound harsh. Try zooming in around the right side of the graph, dragging right-hand limit beyond the rolloff point of the measurements, and create a target curve where the treble tapers down very slightly to the limit, rather than shooting up. Find the shape that sounds detailed and natural, but without harshness or fatigue. This side of the curve can be very sensitive to small changes, and the sweet spot varies depending on speaker and room.

 

Are you suggesting having a slope like this one at the low end ?

 

[rcohen]

Interesting. Are you talking about the left end of the target curve ? To me it seems to follow the measured response ?

I'm basically talking about rounding out that left upper-left corner on your first graph a bit, to see if it improves your impulse response.

 

The idea is that a filter is getting generated based on the difference between the measured curve and the target curve. Filters with high amplitude steep slopes may introduce ringing. Sometimes, though, this ringing can cancel out ringing from your room, and actually fix ringing from the listening position.

 

Either it will sound better, worse, or you won't be able to tell the difference, and you should go by that.

[rcohen]

Regarding the BBC dip, I think that depends heavily on speakers.

 

It's probably best to use the standard response of the speakers as a guide, but you can go with whatever you prefer.

 

I prefer it without the dip, myself.

[rcohen]

BTW, this is another example of Dirac spiking the treble on the right side of the graph. It creates a FR with a nice tilt, but that puts the measured treble over the line and spikes it where the filter rolls off. This sounds edgy and fatiguing to me. You can flatten it out by zooming in, but it's strange default behavior for a speaker with good treble extension. Why not a shelf filter?

 

Regarding clipping, a Clip Protection option similar to the one in JRiver would be nice. That automatically scales down the volume when clipping is detected. No clipping is really what you want, but it's easy to get clipping and bad sound if you're not paying attention.

 

It sounds like it's not so straightforward on Macs.

 

Too bad everything doesn't standardize on double-precision floats. I've actually been struggling a lot with clipping in different parts of my system, especially since introducing a some high efficiency speakers to the mix.

[LBob]

LBob, a couple suggestions based on my experiments. YMMV, so see if this works well for you:

 

1) On the left side, avoid that sharp right-angle change in slope between the measured response and the target curve. The difference between the two curves is the correction filter. When the correction filters have steep slopes with high magnitude, that causes ringing. This is most audible with the bottom frequencies as blurring out fine details. Instead, try to smoothly transition from the measured response to the target curve. This may help with the impulse response.

 

2) On the right size, when the treble angles up like that, it can sound harsh. Try zooming in around the right side of the graph, dragging right-hand limit beyond the rolloff point of the measurements, and create a target curve where the treble tapers down very slightly to the limit, rather than shooting up. Find the shape that sounds detailed and natural, but without harshness or fatigue. This side of the curve can be very sensitive to small changes, and the sweet spot varies depending on speaker and room.

 

Thanks for the suggestions. And thanks again for the utility that converted the 0 degree file to 90 degrees.

I'm a little confused though. I thought that everything to the left of the gold dot in the bass wasn't changed and similarly everything to the right of the gold dot in the high frequencies. Am I mistaken?

[flak]

Flavio,

 

I would be very interested in the post on interpreting impulse response graphs. I tried to google it but I couldn't find anything at a non-engineer level.

 

Hi LBob, I'm reposting that old post with the image which got lost in the original thread:

 

When we look at the improvements produced by digital room correction systems we usually look at the frequency response graphs... or at least this is the aspect most people pay attention to.

 

But in evaluating the fidelity of what we are actually listening to (speakers and listening room as a whole) I believe we should not oversee for example this comments from the REW developer's site (lots of quality info there):

 

"The frequency response is only half of the description of what the system is doing to signals that pass through it, the phase response is the other half. Trying to understand systems by looking at the frequency response alone is like trying to understand a book by reading only the even numbered pages. To really understand you need to look at both. That is a bit problematic, however. The frequency response is fairly easy to understand, but the phase response doesn't give up its secrets quite so easily"

 

The easiest way to evaluate the phase response is to look at the impulse response. I'm posting here a screenshot by a forumer with the pulse responses "before" and "after" correction in his listening room, together with an explanation on how to interpret them by Jakob Agren, one of Dirac research engineers:

 

 

"The impulse response tells us how a system behaves when excited by an impulse. An impulse in this context refers to a signal that in theory has infinite height and no width, that is, all energy comes at the same point in time, with nothing happening before this moment, nor after. For a system consisting of a speaker in a room to be "perfect" it need to do nothing to the signal, we want to hear exactly what is on the recording, nothing more, nothing less. To achieve this, the impulse response needs to be an impulse.

 

If it is not, something has happened to the signal along the way. Worth noting is that there are no systems with a perfect impulse response, since it would (among other things) require infinitely high frequencies to achieve.

To judge if an impulse response is better or worse than another when comparing two different systems, or in this case, the same system before and after compensation, we need decide if the impulse response looks more like an impulse after compensation, or less.

 

A better impulse response will have a more distinct first peak, with more distinct meaning the main impulse is higher compared to the tail than before. We also want the main impulse to be narrower. In the example plot we can see that the ratio of main impulse to tail has increased by about a factor three, the main impulse is higher while the tail is roughly the same. The interpretation of this is that the direct wave (the main impulse) from the speaker to the listener is more distinct than the reflections (the tail) after compensation.

 

The values on the x-axis are time in milliseconds. The delay of the compensated impulse response is a consequence of the compensation. In order to improve the impulse response you need to have some headroom in time in which to apply the compensation.

In order to achieve this, the output is delayed slightly, introducing a latency in the audio chain. This is a static delay, and is not system dependent, meaning it does not depend on the room or the speaker (the delay to the uncompensated impulse response depends on the distance from the speaker to the microphone however).

 

The values on the y axis are normalized and are used to compare the before and after results. In this case we can see that the direct wave from the speaker is much more distinct, as it is about three times higher while the reflections are on roughly the same level. Note that as the y values are normalized, the absolute values have no meaning.

 

This is certainly not all information that is available in an impulse response, but they are hard to inspect visually and most of the information are easier to interpret in the frequency domain"

 

That was it

Flavio

[LBob]

Flavio,

 

Thanks for the plain english explanation of impulse response. Much appreciated. In my case it looks like the impulse response was improved quite a bit. It does look like there is a lot of ringing though especially about 12 ms out. Is that correct?

[Fitzcaraldo215]

I completely agree with most all of LBob's sonic observations about Dirac Live at the beginning of this thread. I am in the process of converting my system to pure PC/JRiver playback, eliminating my Integra prepro, Oppo player and cable box. I had been an Audyssey user with Pro for 6-7 years.

 

I enlisted 5 other good, experienced listeners to help with my trial evaluation of Dirac and an ExaSound e28. We compared the following with hi rez Mch classical recordings from the PC hard drive:

 

 

1. My Integra 80.2+ Audyssey XT/32 Pro calibrated via HDMI (my base system)

 

2. Integra via HDMI with Dirac Live at 88k (DSD sources) or 96k (BD-A sources) sampling

 

3. Dirac Live to e28 via USB, same sampling rates.

 

We were almost completely unanimous about how it sounded and strongly preferred 3. over 2., with 1. least preferred. Most notable, as in LBob's commentary, were the spatial advantages, soundstage depth, apparent body and size of the instruments plus greater smoothness and transparency. The e28 really kicked this up several notches over Dirac + Integra alone. At times, it sounds almost holographic in imaging. Beautiful!

 

One of the listeners has also purchased Dirac to supersede Audyssey into his Marantz 8801 based on what he heard and completing his own test with Dirac Live.

 

incidentally, only the stock Dirac target curve has been used.

 

So, my decision was easy to make in keeping Dirac Live and the e28. I am now in the final stages of going prepro-less in favor of the PC/JRiver. My sound has never been better.

[rcohen]

I completely agree with most all of LBob's sonic observations about Dirac Live at the beginning of this thread. I am in the process of converting my system to pure PC/JRiver playback, eliminating my Integra prepro, Oppo player and cable box. I had been an Audyssey user with Pro for 6-7 years.

 

I enlisted 5 other good, experienced listeners to help with my trial evaluation of Dirac and an ExaSound e28. We compared the following with hi rez Mch classical recordings from the PC hard drive:

 

 

1. My Integra 80.2+ Audyssey XT/32 Pro calibrated via HDMI (my base system)

 

2. Integra via HDMI with Dirac Live at 88k (DSD sources) or 96k (BD-A sources) sampling

 

3. Dirac Live to e28 via USB, same sampling rates.

 

We were almost completely unanimous about how it sounded and strongly preferred 3. over 2., with 1. least preferred. Most notable, as in LBob's commentary, were the spatial advantages, soundstage depth, apparent body and size of the instruments plus greater smoothness and transparency. The e28 really kicked this up several notches over Dirac + Integra alone. At times, it sounds almost holographic in imaging. Beautiful!

 

One of the listeners has also purchased Dirac to supersede Audyssey into his Marantz 8801 based on what he heard and completing his own test with Dirac Live.

 

incidentally, only the stock Dirac target curve has been used.

 

So, my decision was easy to make in keeping Dirac Live and the e28. I am now in the final stages of going prepro-less in favor of the PC/JRiver. My sound has never been better.

Did you rerun the measurements when comparing the e28 and the Integra?

 

I typically find that the differences between DACs and amps disappear when I rerun measurements and room correction on them.

 

Also, sometimes the differences between microphone positions will cause a greater difference than the DACs themselves.

 

I assume you aren't using subs?

[rcohen]

Thanks for the suggestions. And thanks again for the utility that converted the 0 degree file to 90 degrees.

I'm a little confused though. I thought that everything to the left of the gold dot in the bass wasn't changed and similarly everything to the right of the gold dot in the high frequencies. Am I mistaken?

 

There's a crossfade region between the gold dot and the orange dot. After the orange dot, nothing gets changed.

 

Still, if everything to the left of the orange dot gets lowered, that's the same as a spike to the right of the dot. The tilt on the default curve for speakers with extended treble past 20khz creates this condition.