Using AI to understand what I'm hearing and find my perfect sound

[josiah]

Hey everyone,

I recently got some new IEMs and started tinkering with my SMSL DL200 settings. I wanted to share my experience because I did something a bit different: I used AI to help me put what I was hearing into words. It really helped me understand my own preferences and finally nail down my preferred sound signature.

The Test
I used some "torture test" tracks for treble, specifically 4Minute (Crazy), Hyuna (I’m Not Cool), and Kylie Minogue. These songs have sharp synths and high notes that usually wear my ears out fast.

What I found out about the filters

• FL3: This was the "most accurate" setting, but it sounded too shouty and sibilance was high for me. On "Crazy," the sirens were harsh and the girls sounded like they were struggling with the high notes. It actually kind of hurt to hear.
• FL5: This was my favorite. The shoutiness in Hyuna’s chorus sounded fuller and more natural. It didn't sound harsh, but it still felt accurate to me.
• FL6: The same chorus sounded tinny and less full than FL5.
• FL7: The bass was louder but felt loose. The high notes were also a bit irritating.

How AI Helped
The most helpful part was using AI to explain these differences. I described the vocals as feeling "pinched" or like the singers were "struggling," and the AI helped me realize I’m actually sensitive to "pre-ringing," which is a digital artifact.

By talking it through, I was able to identify that I prefer a warmer, yet neutral sound signature. Once I had the words for it, I could dial in my settings with confidence instead of just guessing. If you’re struggling to describe why a piece of gear doesn't sound "right," I highly recommend using AI to help you translate what your ears are hearing.

The Firmware Trap
I also learned a hard lesson about updates. I updated my DL200 to the latest firmware thinking newer is always better. But the update actually changed what the filters do, and my setup felt more fatiguing than before because the mapping had shifted.

My Takeaways

1. You don’t always need to update your firmware if things are working and you like the sound.
2. If you do update, you have to listen to everything again from scratch because the "rules" might have changed.
3. Don't be afraid to use AI as a tool. It helped me realize that "Reference" settings aren't always the best for my biology.

Settling on FL5 made my music fun again. If you're feeling ear fatigue, definitely play around with your filters because the default isn't always the best for your personal taste. I hope my learning experience can help someone else out there!

 

[DVDdoug]

Usually DACs can convert digital to analog better than human hearing and usually those filter settings operate above the audible range and have minimal effects in the audible range...

So I tend to be highly skeptical unless you've done controlled-blind listening tests.

Of course AI has never heard anything and it's "learning" the usual "audiophile nonsense" from the Internet. (Ringing and pre-ringing are real but it shouldn't be audible.)

What is a blind ABX test?.
Controlled Audio Blind Listening Tests (video)
Audiophoolery discusses the FEW REAL characteristics of "audio quality" so you can avoid words like "pinched" or "tight", and ignore all of the other meaningless-vague words that "audiophiles" often use. Ringing and pre-ringing fall into the "time base errors" category but it also affects frequency response. Of course the filtering also affects the frequency response directly, but at very-high or ultrasonic frequencies. The filters have zero effect at low-bass frequencies.

 

[Jim Taylor]

If you’re struggling to describe why a piece of gear doesn't sound "right," I highly recommend using AI to help you translate what your ears are hearing.

I hope that you realize that interaction with AI is a two-way street. AI will introduce possibilities, and then your further interaction with it (what you call "talking it through") will lead it to offer more of what interests you. You, OTOH, will believe that AI understands your problem, leading you to hear whatever panacea AI offers. What happened is that AI manipulated your biases.

Everyone is susceptible to bias. The best way to control (but not eliminate) bias in audio is the double-blind test that @DVDdoug mentioned. Here is a list of biases ...

List of cognitive biases - Wikipedia

... and here is a video by Amir showing how to go about a DBT and why:

Audio Blind Testing - You Are Doing It Wrong! - YouTube

 

[josiah]

Counterpoint: What do you have to lose?

I understand the skepticism, but my experience actually started because the sound felt "off" before I even knew what the settings were. After a firmware update, the menu on my DL200 was in Chinese. I couldn't read the interface, but over time I found myself preferring my gaming headsets (the Fractal Scape and Audeze Maxwell) over my previous daily drivers, the Edition XS and Kali IN-5, because the latter had become too fatiguing.

I spent time testing specific tracks like Higher by Kylie Minogue and Crazy by 4Minute, writing down exactly what I heard. On certain settings, the vocals sounded strained or tinny, while others like FL5 sounded much fuller. I also found that FL7 had more bass, but it didn't feel as tight as I preferred.

I didn't involve AI until after my notes were finished. I used it to help translate and match technical terms to the specific sounds I was already experiencing. The AI didn't lead me to a conclusion; it simply helped me explain why the settings I had already chosen by ear were making such a difference. I went from not being able to stand my favorite equipment to enjoying it again, which is what matters most to me.

I was just trying to share what helped me in case it could help others. I’m not sure why my experience was immediately dismissed, but for me, this exercise was enlightening and actually made my gear usable again.

 

[RexrothPigeon]

On certain settings, the vocals sounded strained or tinny, while others like FL5 sounded much fuller. I also found that FL7 had more bass, but it didn't feel as tight as I preferred.

These were sighted tests, though, so they weren't really "listening tests" since our hearing is subject to predispositions, biases, etc. unless we conspire to circumvent those natural and unavoidable factors.

 

[Keith_W]

Counterpoint: What do you have to lose?

What do you have to lose? You end up going on a wild goose chase whilst being lead by a hallucinating AI.

Use AI as a glorified search engine. I once asked AI a question for which I did not know the answer. It gave me a confident sounding answer. So I asked it for a source. It linked me back to a post I made on ASR. So yeah, thanks AI.

 

[voodooless]

This is what my AI said about those filters:

In practice, most listeners can’t reliably tell the difference in blind tests — the effects sit mostly above ~16-20 kHz or only show up in impulse response rather than tonal balance.

So now what?

I didn't involve AI until after my notes were finished. I used it to help translate and match technical terms to the specific sounds I was already experiencing. The AI didn't lead me to a conclusion; it simply helped me explain why the settings I had already chosen by ear were making such a difference.

No it doesn’t. It just regurgitates the same audiophile nonsense you can find anyway on the internet. It just wants to please you and presents you with a plausible explanation.

We should really build an audio AI agent that gives actual correct answers to these kind of questions. It seems AI output is more trustworthy than an ASR community full of subject matter experts

 

[josiah]

Update: Out of curiosity, I did a blind test. My fiancé randomized the filters so I wouldn't know which was which, and I still reached the same conclusion. Can most people really not notice the difference? Is this like the divide between people who can and can't hear the difference in lossless audio?

 

[OldHvyMec]

but my experience actually started because the sound felt "off" before I even knew what the settings were.

There is a big reason for irritability and human interaction, and it doesn't have to be in the normal hearing range. When you boost frequencies above 15,000 to a point where bugs, rats, and mice are chased from a structure, usually, people can't hear a thing, but they sure do get irritated over extended periods of time. BOOST that to a level you can actually hear, and there will be fighting in the middle of the streets.

It's one of the main reasons valve amps have a settling effect on many people, and they usually don't even notice it happening. I used to play music with my new AB SS amps over long periods of time with my kids. Just random tunes they seemed to like, Beach Boys, old Beatles, CCR, and the older pop music of the 50s, 60s, and early 70s, they would be bouncing around, acting crazy. Then I would swap to my tried and true MC240s. In about an hour, they would calm down a bit, but not from being worn out. If you went back to playing the same music with the SS amps (which happened to be James B's designs with ZERO feedback), the craziness would set back in. The amp measures flat as a board, but its ability to handle and reproduce HF in a transiant piece of music is the difference. My valve amps and class Ds roll off a bit in the higher frequencies; some SS amps actually hold flat and true or even tip up under a heavy load.

I've had guys tell me to measure it, but I really didn't see the point if I had a solution to fix the issue, and really didn't see a point to EQ 15KHz or >

We found that out on 2-cycle Detroit engines vs 4-cycle Cat, Cummins, or Mac engines in Grayhound buses used on long hauls, over 200-2000 miles in the 60s-70s. They went to Cummins, and some Mac pancakes, and the customer complaints went WAY down, not to mention 2-cycle Detroits being horrible polluters and fuel sucking pigs to boot. Most of the buses that the Hound had that were Detroit were sold into city service untill they were phased out completely by polution standards.

What we FEEL and can't perceive as sound (very low sub or ultra high frequencies) can often effect how and what we react to, in a given situation. I have a neighbor right now with a dryer vent from HELL. I don't notice it untill I'm getting short with the Wife or the poor mailman that it effect the same exact way. He ask me how I can stand it when he delivers my mail. I asked him to deliver that same bit he noticed to the neighbor but the neighbors are alway in such a foul mood he just lets it pass. I wonder why the neighbors are in such a foul mood all the time? Maybe the bugs dying from the dryer vent and falling in their Campbell soup has something to do with it.

Regards

 

[voodooless]

AI helped me realize I’m actually sensitive to "pre-ringing," which is a digital artifact.

You’re not.

Audiophile Myth #260: The Detestable Digital Filter Ringing and Real Music...

A blog for audiophiles about more objective topics. Measurements of audio gear. Reasonable, realistic, no snakeoil assessment of sound, and equipment.

archimago.blogspot.com

Also note that likely the filters are mislabeled in the manual:

SMSL filters mislabeled on ES9039 DACs

Several SMSL DACs using versions of the ESS ES9039 chip have mislabeled filters. First, I will use the case of the DL200 using one ES9039Q2M chip to show this is the case and correctly identify the filters. Then, I will show that other SMSL DACs using ES9039 chips namely the D-6S, DO100 Pro and...

www.audiosciencereview.com

 

[Mort]

If you’re struggling to describe why a piece of gear doesn't sound "right," I highly recommend using AI to help you translate what your ears are hearing.

Sorry. I'm gonna need more evidence. This is the same old audiophoolery dressed up in AI.

 

[Klonatans]

What do you have to lose? You end up going on a wild goose chase whilst being lead by a hallucinating AI.

Use AI as a glorified search engine. I once asked AI a question for which I did not know the answer. It gave me a confident sounding answer. So I asked it for a source. It linked me back to a post I made on ASR. So yeah, thanks AI.

It depends. For instance, the level of hallucinations of Perplexity is significantly lower that in the case of ChatGPT.

As to DAC filters I have never been able to spot any difference between them, their impact on FR is probably too minor to be audible, at least for me.

 

[Oddball]

I'd rather ask my completely clueless wife that AI . Level of hallucination would probably be equivalent though, just coming from different angles.

 

[Peterinvan]

This article suggests that corporate leaders will use AI to replace the most expensive workers first. This will backfire….

AI companies will fail. We can salvage something from the wreckage | Cory Doctorow

AI is asbestos in the walls of our tech society, stuffed there by monopolists run amok. A serious fight against it must strike at its roots

www.theguardian.com

 

[mhardy6647]

This article suggests that corporate leaders will use AI to replace the most expensive workers first. This will backfire….

AI companies will fail. We can salvage something from the wreckage | Cory Doctorow

AI is asbestos in the walls of our tech society, stuffed there by monopolists run amok. A serious fight against it must strike at its roots

www.theguardian.com

It would make the most sense, on a value-added basis.

 

[voodooless]

This article suggests that corporate leaders will use AI to replace the most expensive workers first. This will backfire….

AI companies will fail. We can salvage something from the wreckage | Cory Doctorow

AI is asbestos in the walls of our tech society, stuffed there by monopolists run amok. A serious fight against it must strike at its roots

www.theguardian.com

I’m not sure how you reach that conclusion. Actually the opposite is advocated. The target are the replaceable jobs, so usually juniors, ops, clerks, first line support, etc. Not the high paid, not easily replaceable jobs.