Help me create a personalized EQ tool

[markanini]

I've been envisioning a tool that uses an iterative blind approach, to provide a personalized playback EQ.

The user provides a familiar song
The tool generates a playlist with 5-6 randomized EQ variations, including original.
The user listens to each variation, all are unlabeled, and ranks them before confirming their preference.
Finally the the highest ranked EQ is revealed with along with parameters, which the app can retain as a new "original" for future tests.

This could be expanded and refined in many ways, but a basic implementation would be plenty useful for personal use I think. Later the tool could ideally be improved in areas of functionality, like loudness matching, better deterministic controls for EQ variations, averaging between multiple songs.

I'd appreciate any pointers on how to get started.

 

[Berwhale]

I don't think the concept of 5-6 truly randomised EQ variations makes any sense. This is the sort of input you might give to a genetic algorithm, not a human who is going to get bored very quickly.

 

[markanini]

I don't think the concept of 5-6 truly randomised EQ variations makes any sense. This is the sort of input you might give to a genetic algorithm, not a human who is going to get bored very quickly.

The average humans proclivity to boredom is something difficult to solve I'm sure. Do you think something can be done to retain someone that opted in at the very least?

 

[Sean Olive]

I've been envisioning a tool that uses an iterative blind approach, to provide a personalized playback EQ.

The user provides a familiar song
The tool generates a playlist with 5-6 randomized EQ variations, including original.
The user listens to each variation, all are unlabeled, and ranks them before confirming their preference.
Finally the the highest ranked EQ is revealed with along with parameters, which the app can retain as a new "original" for future tests.

This could be expanded and refined in many ways, but a basic implementation would be plenty useful for personal use I think. Later the tool could ideally be improved in areas of functionality, like loudness matching, better deterministic controls for EQ variations, averaging between multiple songs.

I'd appreciate any pointers on how to get started.

We filed a patent on this idea a few years ago. There are different approaches but this is one.

US20230403511A1 - Multistep sound preference determination - Google Patents

Techniques for sound preference determination include performing a first round of option selection by: determining a first set of options for setting one or more audio parameters; for each first pair of options in the first set of options: outputting a demonstration audio content with the one or...

patents.google.com

 

[thewas]

We filed a patent on this idea a few years ago. There are different approaches but this is one.

US20230403511A1 - Multistep sound preference determination - Google Patents

Techniques for sound preference determination include performing a first round of option selection by: determining a first set of options for setting one or more audio parameters; for each first pair of options in the first set of options: outputting a demonstration audio content with the one or...

patents.google.com

Can you share an estimate when such will be marketed in a product?

 

[LTig]

Can you share an estimate when such will be marketed in a product?

I thought one must prove that the patented device actually works before getting the patent approved, hence a product should exist.

 

[thewas]

I thought one must prove that the patented device actually works before getting the patent approved, hence a product should exist.

From what I have heard in the past a working prototype can be also sufficient but I am not a patent lawyer so I might be wrong.

 

[LTig]

From what I know a working prototype can be also sufficient but I am not a patent lawyer.

Neither am I, and maybe in the US you can get a patent for the plain idea.

 

[Keith_W]

There is also David Griesinger's DG Sonic Focus, available for free. The idea is to use your eardrum as a subjective microphone - thus EQ'ing not only your HRTF but everything in the hearing path. It works by making the user match the volume of two test tones across different frequencies. At the end of the process, you should have a personal EQ. The app can be installed as a VST in the signal pipeline, where it automatically applies the EQ.

I have tried it and I had awful results. I think I might be bad at subjectively matching test tones, not sure if it's only me or everybody. The first sign that something was amiss was that I was making -10dB cuts and boosts to my DCA Stealth headphones using the app. You know, the headphone that is supposed to be one of the most neutral. I thought that was really odd, but I kept going. Sure enough, the result was unlistenable. Maybe I was having an off day but I repeated it a couple of times and could not get consistent results. So the app sounded great in theory, but in practice I found it very difficult to volume match test tones.

You may be better at it than I am, so give it a try and report back.

 

[terryforsythe]

I thought one must prove that the patented device actually works before getting the patent approved, hence a product should exist.

There is no such requirement, at least in the U.S. However, "[t]he specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention." 35 U.S.C. § 112(a).

See, also, M.P.E.P. § 2163(I), which states "[t]o satisfy the written description requirement, a patent specification must describe the claimed invention in sufficient detail that one skilled in the art can reasonably conclude that the inventor had possession of the claimed invention" and "[p]ossession may be shown in a variety of ways including description of an actual reduction to practice, or by showing that the invention was 'ready for patenting' such as by the disclosure of drawings or structural chemical formulas that show that the invention was complete, or by describing distinguishing identifying characteristics sufficient to show that the inventor was in possession of the claimed invention." (Citations omitted)

 

[Berwhale]

Given that the circle of confusion is a thing, does a personal EQ even make sense? Would it not be better to train oneself to be more tolerant of broader range of frequency responses in a wider range of conditions? It seems to me that pursuing a personalized EQ is going to lead to an ever increasing number of EQ settings tuned for individual tracks, genres, equipment, listening conditions and even moods. This feels like a recipe for lessening our listening enjoyment, rather than increasing it.

 

[markanini]

Top notch input from Sean Olive! I will digest the patent in the following weeks.

I still very much appreciate any suggestions on getting started with making a basic app. I'm not picky about platforms or languages. As long a it's something a novice can make given a fighting chance and a steady stream of spare time.

 

[Keith_W]

Given that the circle of confusion is a thing, does a personal EQ even make sense?

Of course it does. Your HRTF modifies the sound, so a pair of headphones will perform differently on you than it does on another person.

 

[Sean Olive]

I thought one must prove that the patented device actually works before getting the patent approved, hence a product should exist.

Where did you ever hear that? Certainly not in the USA. There are thousands of patents that exist without products that implement them.

 

[Sean Olive]

Can you share an estimate when such will be marketed in a product?

We have implemented something called Personifi in a JBL headphone app that applies EQ based on a hearing test and user input related to age, gender and listening experience. I talked about this in a couple of slides at my CanJam NYC 2024 talk.

 

[Pe8er]

Love the idea. I think a crude but functional MVP is possible with a little shell scripting on MacOs or Linux. Windows probably too, I'm just not familiar with the platform at all. Just thinking it through:

(Apologies in advance @Sean Olive , I hope I'm not overstepping here!)

- Music source: a folder of music files provided by the user. Allow the user to pick one file per ranking session.
- EQ can be handled by CamillaDSP
- EQ variation generation, unlabeled playback and ranking UI can be handled by command line inputs with basic ASCII GUI
- EQ variation generation and everything related to it, beautifully designed in @Sean Olive's patent, could be handled by script logic too.

All of it is a bit above my pay grade, but I'm pretty sure it could be cobbled together

 

[Berwhale]

Of course it does. Your HRTF modifies the sound, so a pair of headphones will perform differently on you than it does on another person.

I get that using sounds generated in a controlled fashion (i.e. tones and sweeps), but the OP's premise is that the EQ is evolved by listening to a 'familiar track' (which will be very much subject to the circle of confusion) with random EQ applied (which is just adding more confusion). It seems like the Monte Carlo method of EQ design to me.

 

[Tom C]

We have implemented something called Personifi in a JBL headphone app that applies EQ based on a hearing test and user input related to age, gender and listening experience. I talked about this in a couple of slides at my CanJam NYC 2024 talk.

I am very happy with my Personi-fi enabled JBL Tour Pro 2 ear buds.