AutoEQ for speaker and headphone

[pierre]

Here are some updates for V0.5:
1. you can track progress here on this GH project
2. progress

• d3rs : implement most of d3js to make low level plots
• px: high level plots similar to plotly express based on the previous one
• ui-kit: a set of ui components to make the main app less painful to write
• math library is progressing with sparse solver that run in // with good preconditionners ; not super fast yet
• BEM and FEM are working with various type of boundary conditions
• main app:
  • audio player is working well
  • plugins are working, still struggling with a nice to look at interface
  • headphone autoEQ is working
• AUv3 plugins are under way to be able to use and test them in a real DAW: working in Reaper but not yet in Logic

3. what's next

• add roomEQ to the app
• add recording to the app
• polish the plugins look&feel
• remove the small but annoying small bugs

The other features go to the next release after I have finish the previous one: look into V0.6 features

• some are progressing the FEM code and the in-room measurements stuff (for rectangular rooms only)

 

[Philipp Paul Klose]

Will you release on GH or must I compile myself?

 

[pierre]

Will you release on GH or must I compile myself?

I will add an automatic build pipeline soon. Compiling is easy on mac and linux. Untested on windows so likely complicated today.

 

[Philipp Paul Klose]

Nice, lemme know.

 

[pierre]

For once, real progress:
- autoeq: can now do MSO and DBA, iir and fir are supported, mix mode too in order to limit latency
- the player was supposed to be a toy but it is getting full of features and it pretty usable
- the UI is progressing nicely, the plugins still need to be properly connected
Target date for a first major release is around end of year since I have some vacation and should be able to put more
time into it.

You can look at the list of ideas for the next version (2026), i have one million ideas but a few of them are close to
work:
- echo cancellation
- use ml to split the tracks back into stems and use the information to make a better upmixer
- directivity control ala glm (which I am testing today)

A few things are not working yet:
- easily building audio plugins (AUv3 is for now resisting against the rust ui)
- the HAL driver under macos is still not working super well (equivalent of soundflower but working on arm macos)
and of course I have 1 millions small bugs to fix. Design also need tightening, more consistency and more precision.
- windows port is lagging behind and I need a more powerfull machine (my Xmas present that should land soonish)

P.

P.S. If you have some skills in design and software development some help would be welcome.

 

[SlowJim]

This is fantastic, played around with the 0.44 GUI version on Mac. It's far more advanced than my needs but after a bit of tinkering I think I got the hang of it. Thanks pierre!

 

[SlowJim]

If looking for more speaker data you could also add https://audiore.kr/ and others that Spinorama includes, this would open up the list of available speakers significantly.

 

[pierre]

If looking for more speaker data you could also add https://audiore.kr/ and others that Spinorama includes, this would open up the list of available speakers significantly.

- You should definitively have access to all the spinorama.org speakers and measurements (sometimes a lagging a bit behind).
- For audiore.kr they send me their data from time to time. Ping them if you wants more.

 

[pierre]

I am working on the build scripts:

- macOS is working DMG is available
- linux (deb and appimage are available)
- windows soon (or not)

you can download it here
https://github.com/pierreaubert/sotf/releases

 

[SlowJim]

Installed the new 0.5 version (Mac) and am a little confused. How do I get to the speaker selection and auto eq generation?

 

[Philipp Paul Klose]

v0.5.7 runs on my Mac. How do you prefer to receive bug reports?

I let sotf run during dinner and it hogged RAM — 52GB after an hour or so I doubt that so much memory is necessary

 

[pierre]

Installed the new 0.5 version (Mac) and am a little confused. How do I get to the speaker selection and auto

you have 2 menus, 1 down (called Studio Rack maybe not the most explicit title) and 1 in the macos bar

- Library: audio player
- Studio rack: a tool to add plugins in a linear way (one plugin feed the next one) similar to the old hardware audio rack
-> gain -> eq -> level meter for example

- Studio full: same idea but with a graph of plugins to do crossovers and more complicated things
- Recording: record with a microphone your speakers in room
- RoomEQ: use the previous recording to propose an eq
- HeadphoneEQ: use a measurement of a headphone (like the one Amir is doing on ASR) and use it to do a first EQ. Modify to taste
- Spinorama: take a measurement of a speaker with a spinorama and generate an EQ from that

P.

 

[pierre]

v0.5.7 runs on my Mac. How do you prefer to receive bug reports?

- https://github.com/pierreaubert/sotf/discussions/116 is the place.
- https://github.com/pierreaubert/sotf/discussions/114 has the current list of know bugs.

I let sotf run during dinner and it hogged RAM — 52GB after an hour or so I doubt that so much memory is necessary

humm of course it does not leak on my machine. I will investigate.

 

[pierre]

- https://github.com/pierreaubert/sotf/discussions/116 is the place.
- https://github.com/pierreaubert/sotf/discussions/114 has the current list of know bugs.


humm of course it does not leak on my machine. I will investigate.

@Philipp Paul Klose i have found a few memory leaks and one is very likely the prb you have seen.

 

[pierre]

Happy new year and new release today https://github.com/pierreaubert/sotf/releases/tag/v0.5.8

Beta quality on MacOS, untested on Linux and Windows.
App is now in hard freeze on features. Only bugs will be fixed.

Quality improvements

• better UI overall esp. the plugins
• lots of small bugs removed
• windows and linux supported, ARM version for now
• critical bugs: various memory leaks, major bug in EQ plugin

What is working today (at least on a Mac)

• audio player
• audio plugin (a growing set with all the basic plugin plus some which are usually not free: upmixer, denoiser, xtc crosstalk- cancellation, pnd=Polyphonic Note Detection)
• you can EQ yourself any speaker on spinorama.org
• you can EQ yourself any headphone if you have the measurement
• you can record your speakers in your room
• you can eq your speakers in your room

What is not working yet?

• graph of plugins is still too buggy to be used
• new features need a lot of testing (see brand new plugin)

New Features

• plugin denoiser and declicker: remove clicks and noise from old records
• plugin xtc aka cross talk cancellation
• plugin pnd: remove wow/flutter (maybe)
• plugin expander: increase dynamics
• plugin mb expander and mb compressor: multi bands version of the expander and compressor

Binaries

• MacOS: working well
• Linux: untested, i generated the build on a server that does not have an audio card ...
• Windows: untested, i generated the build in a vm ...

Testing will happen next week when I have access to more hardware to do proper testing.

 

[SlowJim]

Happy new year, and thanks for the update. For the spinorama section it would be a big help if we could use keyboard input for min and max freq. It's a pain to use the arrows to go up and down.

Edit - a couple of bugs
-Not able to use spaces or numbers when searching for speaker, the field will not accept the input.
-"PEQ model" dropdown seems buggy, receiving this error when choosing Peaks + Highpass.

https://imgur.com/a/1HiIY7n

Also, selecting anything other than the default (PK - All Peak Filters) causes the autoeq to ignore the frequency limits user has set.

And 2 questions:
-With the original version it seemed like running autoeq multiple times with same parameters would give varying results, sometimes even reducing the score. Is there a way to code this so it will automatically find the best possible eq? E.g. run it 10 times in the background and only present the user with the top result.
-What is the expected format for CSV speaker data upload?

 

[pierre]

Happy new year, and thanks for the update. For the spinorama section it would be a big help if we could use keyboard input for min and max freq. It's a pain to use the arrows to go up and down.

Edit - a couple of bugs
-Not able to use spaces or numbers when searching for speaker, the field will not accept the input.
-"PEQ model" dropdown seems buggy, receiving this error when choosing Peaks + Highpass.

https://imgur.com/a/1HiIY7n

Also, selecting anything other than the default (PK - All Peak Filters) causes the autoeq to ignore the frequency limits user has set.

And 2 questions:
-With the original version it seemed like running autoeq multiple times with same parameters would give varying results, sometimes even reducing the score. Is there a way to code this so it will automatically find the best possible eq? E.g. run it 10 times in the background and only present the user with the top result.
-What is the expected format for CSV speaker data upload?

Thank you for reporting bugs @SlowJim

1. various bugs fixed.
2. the optimiser is not deterministic and will converge to different values at each run. I can add more randomness and run multiple times with multiple algorithms. It is rare that it increase the score a lot (remember that +0.6 is the minimum to get a statistically significant difference). So +/- 0.2 is not important even if pleasing to the eye. There is a lot of parameters to play with. I will add a magic button later (ala I am feeling lucky that usually does reasonable well at finding optimal parameters)
3. it depends what you want to do: for headphone that's freq, spl or freq, spl, phase. For room_eq that's a json file which is described in https://github.com/pierreaubert/autoeq/blob/main/autoeq/bin/roomeq/INPUT_FORMAT.md (that's what is generated by the recording process)

v0.5.9 is currently building.