Understanding the state of the DSP market

[dualazmak]

I prefer to eyeball the impulse myself to determine the delays, determine for myself what should and should not be corrected, and so on.

Me too, exactly!
My policy and practice/implementation in my audio project are almost the same as yours.
Just for example, as shown under the below spoiler cover.

Spoiler

- Precision measurement and adjustment of time alignment for speaker (SP) units: Part-1_ Precision pulse wave matching method: #493
- Precision measurement and adjustment of time alignment for speaker (SP) units: Part-2_ Energy peak matching method: #494
- Precision measurement and adjustment of time alignment for speaker (SP) units: Part-3_ Precision single sine wave matching method in 0.1 msec accuracy: #504, #507
- Measurement of transient characteristics of Yamaha 30 cm woofer JA-3058 in sealed cabinet and Yamaha active sub-woofer YST-SW1000: #495, #497, #503, #507

 

[Multicore]

I thought I would make a public service post about DSP,

Thank you.

In this post, why is the Delay box always exactly one sample?

 

[Keith_W]

Thank you.

In this post, why is the Delay box always exactly one sample?

Because that is how it works?

 

[ernestcarl]

Quick and dirty that's how it looks like with a modest correction down low:

Raw is purple,MathEQ is yellow.

View attachment 445699
Unwraped phase after generated min phase


View attachment 445700
Impulse


View attachment 445701
ETC


View attachment 445702
Step response


(WARNING: simple,amateur, single point measurement)

I’m wondering if the yellow trace is windowed as it appears the room has mostly disappeared (see ETC). If not, then there probably is some overcorrection going on.

 

[Sokel]

I’m wondering if the yellow trace is windowed as it appears the room has mostly disappeared (see ETC). If not, then there probably is some overcorrection going on.

No, not windowed at all, you can tell by the rest.
Room is about 80 m², 11 x 7 x 3.2 m, speakers at the long side, no sub, mic at ~2.7 m.
Correction looks like this:

As minimal as possible.

Edit: added ceiling height.

 

[markstein]

Thanks @Keith_W . Just bought @mitchco 's eBook, and will give Acourate another try. Unfortunately this software is so difficult to use...

 

[Snarfie]

No, not windowed at all, you can tell by the rest.
Room is about 80 m², 11 x 7 x 3.2 m, speakers at the long side, no sub, mic at ~2.7 m.
Correction looks like this:

View attachment 445744

As minimal as possible.

Edit: added ceiling height

9 point measurment (vertical placed mic) of 45 years old IMF Compact II monitors with KEF T27 tweeter which sound after 45 years at last totaly balanced using the white flat target curve.

 

[ernestcarl]

No, not windowed at all, you can tell by the rest.
Room is about 80 m², 11 x 7 x 3.2 m, speakers at the long side, no sub, mic at ~2.7 m.
Correction looks like this:

View attachment 445744

As minimal as possible.

Edit: added ceiling height.

The spikes in the ETC are biased towards high frequencies and yet they have mostly vanished post correction. But, your last magnitude before and after plot says only two peaks in the bass were cut. So there is quite a discrepancy.

 

[Sokel]

The spikes in the ETC are biased towards high frequencies and yet they have mostly vanished post correction. But, your last magnitude before and after plot says only two peaks in the bass were cut. So there is quite a discrepancy.

To tell the truth I don't know how MathEQ works.

By the looks and no delay it suggests they are using min filters. They also say that they correct for magnitude and phase but not the specifics and if they correct phase below the preferred curve.
It sure look like it though, and at an aggressive way which I'm no fun at all.

I'll redo everything tomorrow properly, today I just throw the mic in the room just to see the difference. I'm curious too.

 

[Snarfie]

To tell the truth I don't know how MathEQ works.

By the looks and no delay it suggests they are using min filters. They also say that they correct for magnitude and phase but not the specifics and if they correct phase below the preferred curve.
It sure look like it though, and at an aggressive way which I'm no fun at all.

I'll redo everything tomorrow properly, today I just throw the mic in the room just to see the difference. I'm curious too.

Be sure you do a vertical measurment as described on their site.

"- Place the microphone at your ear height. Set it vertically."

 

[krabapple]

Nice read but no in depth coverage of the big commercial players in 'room correction' DSP? (Audyssey, Dirac...) These are what the majority of audio fans will encounter before tools like REW, Acourate, etc.

Is it because there's too much secret sauce in them to comment on them at length?

 

[Keith_W]

Thanks @Keith_W . Just bought @mitchco 's eBook, and will give Acourate another try. Unfortunately this software is so difficult to use...

I think Acourate is easier to use than REW in some ways. Lately I have been experimenting with REW to see if I can do what I can with Acourate. Short answer - you can't. Not without third party software. The two deficiencies of REW: no built-in convolver, and no phase correction.

No built-in convolver means you have to:
- Export your filter into a .WAV file, manually specify the filter length (65536) and the impulse peak (32768).
- Write a config file for your convolver and load it
- Route signal from REW into your convolver. Hopefully it has a WDM input, otherwise you will need ASIO4ALL.
- Because convolvers create additional latency, you need to set up some kind of timing reference. I prefer loopback. That means even more futzing around.
- Check channel assignments with the signal generator before you send a logsweep through it.
- Take your measurement

In other words, there are 3 separate pieces of software in the signal chain, just to take a measurement of an active multichannel speaker.

In Acourate, create your multiXO filter .WAV, load it into your logsweep recorder, and you're done. 2 steps. Audiolense is just as easy, you set up channel assignments then press record.

Then there are the room macros. If what you want to do is very simple, just use the macros. In 10 minutes, you have your filters and you could be listening to music. Those macros are the only automation in Acourate. If you want, you can decide not to use the Macros and execute Acourate's normal functions instead. You will get the same result, but it will take you a lot longer. If you want to replicate that in REW, not only do you have to know what operations to do and in what order, you will also have to export files to rePhase in between steps and re-export them back to REW.

That's the difference between software that was designed from the ground up to create DSP filters, and measurement software which has been pressed into service to make filters. Whilst it is possible, it is inadequate and inconvenient. Don't get me wrong, REW is excellent measurement software. But it has its limitations.

 

[Snarfie]

Nice read but no in depth coverage of the big commercial players in 'room correction' DSP? (Audyssey, Dirac...) These are what the majority of audio fans will encounter before tools like REW, Acourate, etc.

Is it because there's too much secret sauce in them to comment on them at length?

A recent (october 2024) paper about Exploring the Current Landscape of Open Research Software in Room Acoustics from Eindhoven Technical University. Overview in the Abstract.

 

[Multicore]

Because that is how it works?

Idk. My memory of this stuff is hazy.

Say you want to build a FIR sub crossover and the sample frequency is 48k. 24x under sample would put Niquist at 1k, which might be sufficient, and cut the arithmetic workload by 95%.

 

[Robbo99999]

It's a writeup but not for people new to audio, I didn't understand a lot of it and I've been exposed to a lot here on ASR and have a pretty darn good working knowledge experience re implementing EQ in headphones and speakers, so I think it could have benefitted by starting off literally talking about what the purpose of DSP is and how it's fundamentally about changing the frequency response, but this is a deep deep dive into it this writeup and I couldn't read all of it in an informal sitting, so I can't make any more detailed comments about it other than the ones I've made so far as I've not read it all! Must have taken some effort to do though. Not noob friendly. (and not sure that it needs to be made that complicated)

 

[Sokel]

The spikes in the ETC are biased towards high frequencies and yet they have mostly vanished post correction. But, your last magnitude before and after plot says only two peaks in the bass were cut. So there is quite a discrepancy.

Ok, I was too curious to wait and I also tried to eliminate pipeline glitches (measuring from file while trying to preserve timing is tricky) . Mic at 1 m this time .
So I measured everything through foobar as:

Raw: Math bypassed by it's comparing button

Minimal: as shown:

Overcorrection: as shown:

Results (description on chart) :


FR


Impulse



Step



ETC



unwrapped phase

So definitely a pipeline error before, it seems gentle even at way overcorrection.

Edit: I was also able to replicate the previous "roomless" one by capturing and averaging (8 times) the same playback through RTA this time.

 

[SorenTyson]

Most went over my head too, but thanks goes to the OP for the comprehensive guide

Completely over my head Bookmarked and I will re-read! Thanks for posting this.

 

[Snarfie]

Ok, I was too curious to wait and I also tried to eliminate pipeline glitches (measuring from file while trying to preserve timing is tricky) . Mic at 1 m this time .
So I measured everything through foobar as:

Raw: Math bypassed by it's comparing button

Minimal: as shown:

View attachment 445777

Overcorrection: as shown:

View attachment 445778

Results (description on chart) :

View attachment 445779
FR

View attachment 445780
Impulse


View attachment 445781
Step


View attachment 445782
ETC


View attachment 445787
unwrapped phase

So definitely a pipeline error before, it seems gentle even at way overcorrection.

Thanks for the measurements. Qurious what your findings are when listening to music.

 

[Keith_W]

Is it because there's too much secret sauce in them to comment on them at length?

I did not mention Audyssey explicitly in the write-up, but it is a mixed-phase automated DSP solution specifically tailored to run on AVR's. As I mentioned, AVR's have specific DSP requirements - low latency is a must, and computing power is limited. Minphase IIR or mixed phase is a great solution. I have never played with Audyssey since I don't own a TV, let alone an AVR.

As for Dirac, everything I know about it I learned from @mitchco Dirac walkthrough. Mitch is very nice (he's a Canadian after all) and he does not explicitly point out what features are missing in his walkthrough. So I don't know if it can import curves, let you manually manipulate them, create complex crossovers, design all-pass filters, use an MMM measurement, split and join two measurements (for e.g. a ground plane measurement of a woofer + quasi-anechoic measurement of a speaker) and so on. I think the answer is likely to be "no". The biggest question of all, which he does not answer, is whether you can over-ride Dirac's correction if you see it doing something you don't like. Or even if it has the tools to show you that it is doing something you don't like, for example lock on to a reflection instead of the main impulse, and send subwoofer correction the wrong way. There are certainly many reports of wonky subwoofer correction in Dirac on ASR and other forums. It does seem very easy to use, just not particularly powerful or flexible. IMO their asking price is pretty eye popping.

You can do all these in REW and more, but REW isn't as "nice" to use as Dirac. No pretty graphics, no hand-holding, no automation. If you screw up, you screw up. And there are plenty of opportunities to screw up. Dirac tries to stop you from screwing up, but no algorithm can compensate for rubbish measurements.

 

[Chr1]

Couple of questions...

Can you use additional VSTs along with DIRAC?

Does anyone know what the differences/similarities are between Dirac and MathAudio RoomEQ?

Full disclosure : As MathAudio is free, Diracs high price alone is a deal breaker for me!