• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). There are daily reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Introducing Hang Loose Convolver from Accurate Sound

mitchco

Addicted to Fun and Learning
Manufacturer
Joined
May 24, 2016
Messages
586
Likes
2,206
Hello fellow ASR enthusiasts,

Just over a year ago @dallasjustice was kind enough to announce my new DSP calibration service here on ASR. Since then, 125 individuals have used my service to calibrate their sound systems. Coming back to my audio roots after decades in the software industry was fun, and I learned a lot.

I also accumulated independent measurement data from different loudspeakers in different rooms from around the world. Interesting patterns began to emerge, not only from the measurement data, but from client preferences for corrected in-room frequency responses.

Data analysis of the acoustical measurements pointed to a mathematical formula for applying the optimum amount of digital room correction to achieve accurate sound. This is what made (and makes) my DSP calibration service unique.

If interested, I hope to publish the results of the anonymized data in the next edition of my DSP book. Meanwhile, look for an upcoming Master Class video on Digital Room Correction on my website homepage, which will teach you about the process.

The need for a Hang Loose type of convolver began when I was working as a recording/mixing engineer in the 80’s. The band, producer, and I would huddle in front of the mixing console and monitors while I manually switched in/out various “premastering” 2 channel eq before the mix heads off to the mastering lab. It was our final opportunity to determine which eq gave us the best sound quality. It would have been great to have had something that was automated.

Then, 10 years ago, while creating digital room correction filters as a hobbyist, I wanted a convolver that allowed easy level matching and real-time switching between filters so I could determine which one I liked best. I could not find one that would accomplish this however. Ten years later, my DSP calibration clients were asking, how do I tell which filter sounds better? How do I know if it sounds better with or without the filter? So, the impetus for developing Hang Loose Convolver (HLC) had finally begun.

The capability to level match digital correction filters to bypass level and switch seamlessly between filters in real-time has been designed into HLC. There is science behind the design, which can be read here: https://accuratesound.ca/products.html

Key Features:
• Designed for loudspeakers and headphones
• Seamless real-time switching between filters
• Autogain level matching with manual gain adjustment
• Import Acourate, Audiolense, Focus Fidelity, and REW filters
• Supports stereo 32-bit float wav FIR filters in a zip file
• Automatic filter switching based on host sample rate
• 6 Filterbanks x presets = dozens of FIR filters
• System-wide and app specific convolution capabilities
• Zero latency, uniform partition convolution engine
• Standalone application mode and VST3/AU plugin mode

HLC Operations Guide (PDF).

Tested (so far) with:
Roon on Mac and Windows operating in standalone mode with loopback.
JRiver on Mac and Windows using VST3 plugin.
Audirvana on Mac and Windows using AU and VST3 plugins
Qobuz on Mac and Windows using standalone mode with loopback.
HQPlayer on Mac and Windows using standalone with loopback.
System Wide Audio on Mac and Windows using standalone mode with loopback.
Popular DAW’s like Ableton and Reaper on Mac and Windows using AU and VST3 plugins.

Pic of HLC running in standalone mode on a Mac M1:

HLC on Mac.png


I am happy to answer any questions you may have.

Happy listening!

Kind regards,
Mitch
 

Belker

Member
Joined
Feb 20, 2021
Messages
40
Likes
76
Location
Sweden
Looks nice! But what is the big difference between this and just using DSP presets in Roon? The small gap of silence while Roon switches preset?
 

abdo123

Master Contributor
Forum Donor
Joined
Nov 15, 2020
Messages
5,970
Likes
5,902
Location
Brussels, Belgium
Does this convolver work with the OS mixer or does it make its own virtual interface? Edit: I just read that both options are available.

Have you ever considered making your own DSP engine? a feature full convolver seems like a great idea but a little 'meh' when you look at the grand scheme of things. I mean 130$ for a software I would use for a few times and move on?

People are very likely to move on to Roon (or whatever ecosystem they're in) or Equalizer APO for once they're set on which filter set they would like to use.

Furthermore, I feel like for a lot of people use their setups to listen to records (analog sources), or watch TV (optical or ARC) and they need a dedicated engine to do this.

an Intel NUC board can provide plenty of processing power and flexibility. and people can buy the necessary interfaces and/or DACs for their needs.
 

Χ Ξ Σ

Senior Member
Joined
Jul 16, 2020
Messages
390
Likes
1,587
Location
UTC-8
WOW!

This looks nice, Congrats! I hope this also works with SoundSource, Audio Hijack, and Loopback on Mac.
 
Last edited:
OP
mitchco

mitchco

Addicted to Fun and Learning
Manufacturer
Joined
May 24, 2016
Messages
586
Likes
2,206
Looks nice! But what is the big difference between this and just using DSP presets in Roon? The small gap of silence while Roon switches preset?

Thanks! HLC was not designed just for Roon... There is a scientific approach applied that is very specific. Roon's convolver is not zero latency and depending on the length of the filter used, there is 1/2 to 3/4 of a second silence gap when switching filters, which has a negative effect on our echoic memory. Given that echoic memory is short: 1 to 10 seconds, with 3 to 4 seconds being typical, having the switching transition literally "blanked out" impairs our ability to remember the original sound - it is pretty much wiped out. So making decisions on which filter sounds best is impaired in Roon. The real-time switching of the transition between filters is what gives us the immediate clue on how the filter sounds different. Roon cannot do this. This is in addition to not being able to easily level match filters in Roon.
 

pierre

Addicted to Fun and Learning
Forum Donor
Joined
Jul 1, 2017
Messages
714
Likes
1,775
Location
Switzerland
Thanks! HLC was not designed just for Roon... There is a scientific approach applied that is very specific. Roon's convolver is not zero latency and depending on the length of the filter used, there is 1/2 to 3/4 of a second silence gap when switching filters, which has a negative effect on our echoic memory. Given that echoic memory is short: 1 to 10 seconds, with 3 to 4 seconds being typical, having the switching transition literally "blanked out" impairs our ability to remember the original sound - it is pretty much wiped out. So making decisions on which filter sounds best is impaired in Roon. The real-time switching of the transition between filters is what gives us the immediate clue on how the filter sounds different. Roon cannot do this. This is in addition to not being able to easily level match filters in Roon.

lgtm. Does it support large configuration like 9.1.6? What is the latency at the end. The doc shows 2048 samples … plus the time in the convolver etc… if it is just to decide which filter is the best thats fine. If I want to put it on the master bus, latency need to be reasonable.
 

Belker

Member
Joined
Feb 20, 2021
Messages
40
Likes
76
Location
Sweden
Thanks! HLC was not designed just for Roon... There is a scientific approach applied that is very specific. Roon's convolver is not zero latency and depending on the length of the filter used, there is 1/2 to 3/4 of a second silence gap when switching filters, which has a negative effect on our echoic memory. Given that echoic memory is short: 1 to 10 seconds, with 3 to 4 seconds being typical, having the switching transition literally "blanked out" impairs our ability to remember the original sound - it is pretty much wiped out. So making decisions on which filter sounds best is impaired in Roon. The real-time switching of the transition between filters is what gives us the immediate clue on how the filter sounds different. Roon cannot do this. This is in addition to not being able to easily level match filters in Roon.
Not only for Roon, I buy that! The value of zero latency filter switching can be argued, but let’s not go there. But sticking to Roon for a while, almost (good enough) zero delay is achieved by doing a couple of filter sketches in Roon EQ, before going to convolution.
On the topic of level setting, how do you even do that on different EQs? At what frequency? At what average? Personally I find this difficult when comparing EQ settings. Usually you strive to reduce peaks, and the more you reduce them, the more this setting loose in ABX comparation, due to decreased overall loudness. What is your take on this?
 
Last edited:
OP
mitchco

mitchco

Addicted to Fun and Learning
Manufacturer
Joined
May 24, 2016
Messages
586
Likes
2,206
Does this convolver work with the OS mixer or does it make its own virtual interface? Edit: I just read that both options are available.

Have you ever considered making your own DSP engine? a feature full convolver seems like a great idea but a little 'meh' when you look at the grand scheme of things. I mean 130$ for a software I would use for a few times and move on?

People are very likely to move on to Roon (or whatever ecosystem they're in) or Equalizer APO for once they're set on which filter set they would like to use.

Furthermore, I feel like for a lot of people use their setups to listen to records (analog sources), or watch TV (optical or ARC) and they need a dedicated engine to do this.

an Intel NUC board can provide plenty of processing power and flexibility. and people can buy the necessary interfaces and/or DACs for their needs.

HLC is designed to run on any platform and work with any audio source. For Mac, there is a standalone app plus VST3 and AU plugins. On Windows there is a standalone app and VST3 plugin. Shortly will be available on popular distros of Linux, with standalone app and VST3 plugin followed by iOS and Android apps. That's 9 convolvers running on the majority of platforms for a one time perpetual license of $129 with with free upgrades.

HLC is a fully fledged DSP engine. Getting easy to use convolution on the majority of platforms is phase 1 of this release, in addition to it's specialized capability. There will be other "DSP" features moving forward that I am not ready to disclose yet.

I have numerous DSP calibration clients that have turntables, TV's, CD Players, etc. So yes, if you want DSP correction for your loudspeakers and headphones using these sources, you need a convolver.

PS. the convolution engine is highly efficient, so not a lot of CPU is required.
 

abdo123

Master Contributor
Forum Donor
Joined
Nov 15, 2020
Messages
5,970
Likes
5,902
Location
Brussels, Belgium
HLC is designed to run on any platform and work with any audio source. For Mac, there is a standalone app plus VST3 and AU plugins. On Windows there is a standalone app and VST3 plugin. Shortly will be available on popular distros of Linux, with standalone app and VST3 plugin followed by iOS and Android apps. That's 9 convolvers running on the majority of platforms for a one time perpetual license of $129 with with free upgrades.

HLC is a fully fledged DSP engine. Getting easy to use convolution on the majority of platforms is phase 1 of this release, in addition to it's specialized capability. There will be other "DSP" features moving forward that I am not ready to disclose yet.

I have numerous DSP calibration clients that have turntables, TV's, CD Players, etc. So yes, if you want DSP correction for your loudspeakers and headphones using these sources, you need a convolver.

PS. the convolution engine is highly efficient, so not a lot of CPU is required.


Well this is amazing news then! I was kind of getting tired of MiniDSP's almost monopoly of the dedicated DSP engines, and i wanted a solution to apply 65332 taps FIR filters to my records and TV live.

hopefully your software at some point in the future will make this possible with little inconvenience.
 
OP
mitchco

mitchco

Addicted to Fun and Learning
Manufacturer
Joined
May 24, 2016
Messages
586
Likes
2,206
lgtm. Does it support large configuration like 9.1.6? What is the latency at the end. The doc shows 2048 samples … plus the time in the convolver etc… if it is just to decide which filter is the best thats fine. If I want to put it on the master bus, latency need to be reasonable.

Hi Pierre, the convolution engine is 0ms latency. The "host" determines the round trip latency. In the case of the 2048 sample buffer size you saw in the doc was for the standalone "host" in a scenario where overall system latency does not matter. However, the buffer size can be adjusted by the host, but the convolution engine is 0ms latency. If you are using HLC plugin in a DAW, then the DAW determines the round trip latency while the convolution engine is still 0ms latency.
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
37,996
Likes
162,681
Location
Seattle Area
Hi Mitch. I have a question. Is your tool for testing different filters and once one decides on one, you then import it into other convolvers? I am asking as a user of Roon. How would I use the end result there?
 

phoenixdogfan

Major Contributor
Joined
Nov 6, 2018
Messages
2,278
Likes
2,681
Location
Nashville
Hi Mitch, how does your zeroi latency compare with just running Dirac Live or any other DSP program doing its own convolutions as a stand alone or vst. Maybe just compare DL 3 and Audessy for starters?
 

KSTR

Major Contributor
Joined
Sep 6, 2018
Messages
2,075
Likes
4,589
Location
Berlin, Germany
Hi Klaus, not currently, but it is an easy feature enhancement to implement...
Good to know.
Another question, IIRC, VST3 offers 64bit precision (64bit floats, I would assume), is the internal processing (which is most likely FFT/iFFT-based convolution) also 64bit floats?
Not that this would matter for music playback but sometimes it would be nice to have lowest possible artifacts for more academic purposes ;-)
 
OP
mitchco

mitchco

Addicted to Fun and Learning
Manufacturer
Joined
May 24, 2016
Messages
586
Likes
2,206
Not only for Roon, I buy that! The value of zero latency filter switching can be argued, but let’s not go there. But sticking to Roon for a while, almost (good enough) zero delay is achieved by doing a couple of filter sketches in Roon EQ, before going to convolution.
On the topic of level setting, how do you even do that on different EQs? At what frequency? At what average? Personally I find this difficult when comparing EQ settings. Usually you strive to reduce peaks, and the more you reduce them, the more this setting loose in ABX comparation, due to decreased overall loudness. What is your take on this?

Re: zero latency filter switching can be argued... You would be arguing against the science ;-) Also, HLC is not intended for filter sketches, it is designed to host high resolution FIR filters that can contain both frequency and timing corrections.

The level matching procedures are in the Operations Guide. HLC comes with an "Autogain" feature that calculates the DC Gain of the FIR filter and applies the gain to the filter via the gain slider. Depending on the filter's frequency response, this may be enough. However, depending to what degree you want to level match the filters will determine which procedure to use.

Included with the download is a full range Pink Noise .wav file one can play in a music player while using an SLM to level match the filters to bypass (C weighting, slow average). More science incoming, if the filters and bypass are not level matched, then whichever is louder, even by 1 dB SPL, will be perceived as "sounding better" As the actual loudness changes, the perceived loudness our brains hear will change at a different rate, depending on the frequency. Often referred to as the Equal loudness contour.

Simply put: at lower listening volumes mid-range frequencies sound more prominent, while the low and high frequency ranges seem comparatively quieter. At higher listening volumes the lows and highs sound more prominent while the mid-range seems comparatively quieter. Therefore, our ears interpret louder sound to have a fuller bass and more clarity on the top end. Conversely, we perceive quieter sounds to be thinner and duller. Even an increase of 1 dB SPL can create the illusion of a "better" sound. So "easy" level matching of filters and bypass was one design goal for HLC. Finally, with the addition of another person working the controls, HLC can meet the formal ABX testing criteria.
 
OP
mitchco

mitchco

Addicted to Fun and Learning
Manufacturer
Joined
May 24, 2016
Messages
586
Likes
2,206
Hi Mitch. I have a question. Is your tool for testing different filters and once one decides on one, you then import it into other convolvers? I am asking as a user of Roon. How would I use the end result there?

Hi Amir, yes, that is one way to use HLC. Both Roon and HLC take the same correction filter format, including the same filter at different sample rates bundled in a zip file. Other convolvers may require a .cfg file to go with it.

Another way to use HLC with Roon is to loopback the output of Roon into the input of HLC to convolve the audio and output to one's DAC. The connection is lossless as show in the attachment.

In addition to testing FIR filters for preference, whether speakers or headphones, I wanted to create a convolver that is easy use and available on all platforms and to all applications plus system wide audio. This is so folks can easily host high resolution FIR correction filters that go beyond simple PEQ's and experience what a properly designed DSP FIR filter can offer in increased SQ.
 

Attachments

  • Roon Music Player.png
    Roon Music Player.png
    691.6 KB · Views: 128

Another Bob

Member
Forum Donor
Joined
Nov 28, 2018
Messages
71
Likes
117
Location
Madison, WI
If interested, I hope to publish the results of the anonymized data in the next edition of my DSP book.
Hi Mitch. You’ve been teasing us with this for quite a while. Any target date? I’ve been delaying my purchase of Audiolense until it comes out. (And your convolver looks like a very interesting product, BTW.)
 
Top Bottom