Broooo the speeeee you are developing this is insane!!!! I did about 25 tests allready and i love the results they are super interesting! Very happy you introduces ab testing this will be very benefitial! Some things i would be carefull about: Make the music more divers. I tested many of the best mastereed songs of my discography, and while the results were surprisingly similar, every song is mastered with an individual preference by a human so you need variations in material until you find songs that are mastered in a way that they perform similar to a avarage of say hundert great mastered songs. King gizzard is great and they are mastered great but still this will lead to a target curve that is inherently wrong. Even increasing to 6 good masterred tracks that change at randome will improve the process bigtime! 2. Please please please make a mode for general purpose. I want to experiment with this thing, find eqs curves for speakers and masters, and i dont want to ruin your dataset. Make a all purpose mode without headphone modell, that would be so amazing!!!! 3. While i apriciate reducing the trials, i rather go for precission to be honest. I bet a lot of consumers like a fast workflow, but if would personally literally cry if you make it no more possible to make the detailed longer tests like before becouse they worked soooo well allready
Interesting thing I noticed was that with different headphones I always had the same curve, but I could still hear the differences between each headphone and had clear preferences between them. I feel like a driver will always be able to produce any kind of curve using a single swept tone but driver type number and geometry drastically change the ability of a device to have a clear supposition of many different frequencies at the same time and that it changes the character of a sound REGARDLESS of an overall frequency response preference.
I thought you were bluffing - this ACTUALLY works. It has improved everything I've thrown at it. Wow man, just wow. Keep up the good work and good luck with your future research
Remember I (hearingAIDS) have been saying for years in your discord that harman was too hot under 105hz and too hot at ~5k, but too low above 10k? 😁Good job mate. I believed in your curiosity and willingness to learn through the years and you went about the right way to prove it to others after stumbling and getting closer with each attempt. There's still work to be done!
fills my heart that your first official product isn't some cashgrab collab but something actually revolutionary and for FREE, clearly you love audio quality and love working on it, and i love you Sharur😚
I did the test and got a personalized curve, then compared it to the Harman in other tracks, and I liked the Harman better. I think it would make sense to extend the test to load, for example, 4 tracks, or load tracks indefinitely to get a more stable and accurate curve
This is just seriously great for the community! Weather the audio science community or the greater audio community accepts PEQDB's results or not don't matter. This is just a net positive for everyone who enjoys listening to proper sounding music.
I EQ-ed my ATH-R70X using the tests & it has more bass & is more clearer than it was tuned to Harman 2018, Harman sounds muffled in comparison :D The test probably helps get a better sound for a person than just EQ-ing off graphs because it also accounts a headphone unit variation in sound & the loss of hearing a person might have since they test it with their hearing it produces greater results for them. Great work!
This is actually really neat. Awesome work! I'm surprised how close the IEM and OE targets are. I've wondered why the OE target lacked in bass for quite awhile and why barely any headphone has IEM curve bass.
A suggestion, maybe have a score to show if a user is not being consistent across tests or choosing one FR over another when they are actually the same.
Hey, I just tried the A/B test mode and noticed a couple of things. First, while it’s not a major issue, I feel like I shouldn’t have to pause one track and then press play on the other. It would be much more intuitive if pressing play on one track automatically paused the other. This is more of a user experience suggestion, so it’s not a huge deal. More importantly, I set the number of trials to 99, but the first twenty trials sounded nearly identical, so I ended up selecting "no preference" for all of them. I didn’t complete the remaining 79 trials because of this. It seems like the difference between the two samples needs to be much more noticeable for the test to be effective.
I think being able to eq literally any iems or headphones that don't have any sort of graphs or measurements esp in the budget region in third world countries like mine would be an absolute game changer. I don't know if thats even possible but if it is then PEQdb is gonna be the endgame of every audio device.
Good work so far! May I suggest you add more variety in the music clips? I find some parts of the clips weird/unnatural and hard to judge. I would worry about your ultimate target curve being overtrained to the very few preselected clips. I know you can upload your own songs and I don't know the percentage of people who do that. Maybe you could analyze that and see if there is an impact.
@@sharuraudioHow negligible? Could you share some data in an update? Also, it would be interesting to see confidence intervals for the targets, although they may be difficult to interpret depending on how many knobs you're actually fiddling with. The preselected audio, are they from different records? To me they sound like they could be off the same album and hence mastered similarly, as another commenter seemed to imply.
The testing is only with music, right? Maybe people prefer different EQ when watching other content like videos, movies etc.. At least I am pretty sure I do. In my opinion music is more forgiving when it comes to tuning.
Do you plan on adding Soundcore Life Q30 to the headphone list? It was a popular budget ANC wireless OE headphone due to being able to get somewhat close to the harmann OE target with the in app 10 band EQ (oratory1990 has measurements and EQs for ANC off/on).
This is hugely flawed. All you're really doing here is equalizing a headphone to your preference for a single track. It's more of an preference mastering EQ target for that specific track than an overall headphone target. You need to test 5 to 10 different tracks at least and average them out like Harman did; otherwise, the equalization just sounds crap on any track other than the one you used for the A/B test.
You also need to consider the significant variations in Gras measurements. You must ensure that everyone has the same starting point for the test, but you cannot achieve that with the methods you have used. Most Gras measurements by different measurers using the exact same rig are so far off. That's something that Harman controlled for but it's impossible for you to do with this method.
@@geo2160 Hahaha! If only that was a good argument. No, I'm just a mere hobbyist who's has read a lot of papers on this subject. There's a reason this hasn't been done, and it's not because others haven't already thought of it. It's because it's highly flawed. There are much better methods being worked on in audio labs around the world around measuring and simulating HRTFs. SoundID Personalization by Sonarworks did something similar, which is more accurate but still very flawed.
A/B so much simpler and better
A and B sound the same to me 95% of the time. You should mix up more drastic differences.
Yeah I can’t tell
Broooo the speeeee you are developing this is insane!!!! I did about 25 tests allready and i love the results they are super interesting! Very happy you introduces ab testing this will be very benefitial! Some things i would be carefull about:
Make the music more divers. I tested many of the best mastereed songs of my discography, and while the results were surprisingly similar, every song is mastered with an individual preference by a human so you need variations in material until you find songs that are mastered in a way that they perform similar to a avarage of say hundert great mastered songs. King gizzard is great and they are mastered great but still this will lead to a target curve that is inherently wrong. Even increasing to 6 good masterred tracks that change at randome will improve the process bigtime!
2. Please please please make a mode for general purpose. I want to experiment with this thing, find eqs curves for speakers and masters, and i dont want to ruin your dataset. Make a all purpose mode without headphone modell, that would be so amazing!!!!
3. While i apriciate reducing the trials, i rather go for precission to be honest. I bet a lot of consumers like a fast workflow, but if would personally literally cry if you make it no more possible to make the detailed longer tests like before becouse they worked soooo well allready
Interesting thing I noticed was that with different headphones I always had the same curve, but I could still hear the differences between each headphone and had clear preferences between them. I feel like a driver will always be able to produce any kind of curve using a single swept tone but driver type number and geometry drastically change the ability of a device to have a clear supposition of many different frequencies at the same time and that it changes the character of a sound REGARDLESS of an overall frequency response preference.
I thought you were bluffing - this ACTUALLY works. It has improved everything I've thrown at it. Wow man, just wow. Keep up the good work and good luck with your future research
of course it works.. I study and research audio at Stanford University
Very promising start for PEQdb
PEQdb keeps on winning each update 🔥
Remember I (hearingAIDS) have been saying for years in your discord that harman was too hot under 105hz and too hot at ~5k, but too low above 10k? 😁Good job mate. I believed in your curiosity and willingness to learn through the years and you went about the right way to prove it to others after stumbling and getting closer with each attempt. There's still work to be done!
always
fills my heart that your first official product isn't some cashgrab collab but something actually revolutionary and for FREE, clearly you love audio quality and love working on it, and i love you Sharur😚
A/B testing seems like a much simpler method compared to the rating system. Love the way this is better everyday.
I did the test and got a personalized curve, then compared it to the Harman in other tracks, and I liked the Harman better. I think it would make sense to extend the test to load, for example, 4 tracks, or load tracks indefinitely to get a more stable and accurate curve
This is just seriously great for the community! Weather the audio science community or the greater audio community accepts PEQDB's results or not don't matter. This is just a net positive for everyone who enjoys listening to proper sounding music.
Thank you for the video, Sharur
Why are we still here? Just to suffer?
I would be interested in seeing whether the results from A/B mode differ from those in regular mode.
I EQ-ed my ATH-R70X using the tests & it has more bass & is more clearer than it was tuned to Harman 2018, Harman sounds muffled in comparison :D The test probably helps get a better sound for a person than just EQ-ing off graphs because it also accounts a headphone unit variation in sound & the loss of hearing a person might have since they test it with their hearing it produces greater results for them. Great work!
Congrats! Great work!
Have a nice Christmas!
This is actually really neat. Awesome work!
I'm surprised how close the IEM and OE targets are. I've wondered why the OE target lacked in bass for quite awhile and why barely any headphone has IEM curve bass.
A suggestion, maybe have a score to show if a user is not being consistent across tests or choosing one FR over another when they are actually the same.
Hey, I just tried the A/B test mode and noticed a couple of things. First, while it’s not a major issue, I feel like I shouldn’t have to pause one track and then press play on the other. It would be much more intuitive if pressing play on one track automatically paused the other. This is more of a user experience suggestion, so it’s not a huge deal.
More importantly, I set the number of trials to 99, but the first twenty trials sounded nearly identical, so I ended up selecting "no preference" for all of them. I didn’t complete the remaining 79 trials because of this. It seems like the difference between the two samples needs to be much more noticeable for the test to be effective.
totally agree on both point
regarding the UI, this will be addressed after beta. Thanks for the algorithm suggestions
I think being able to eq literally any iems or headphones that don't have any sort of graphs or measurements esp in the budget region in third world countries like mine would be an absolute game changer. I don't know if thats even possible but if it is then PEQdb is gonna be the endgame of every audio device.
PEQdb target is almost Truthear Nova frequency response with 20hz -4db ( personally no need eq for Nova👑)
Pedb is a sick name kinda 🔥
no
@sharuraudio Yea, no xD
How many ABX runs would you say is good enough?
Good work so far! May I suggest you add more variety in the music clips? I find some parts of the clips weird/unnatural and hard to judge. I would worry about your ultimate target curve being overtrained to the very few preselected clips. I know you can upload your own songs and I don't know the percentage of people who do that. Maybe you could analyze that and see if there is an impact.
there's negligible impact
@@sharuraudioHow negligible? Could you share some data in an update? Also, it would be interesting to see confidence intervals for the targets, although they may be difficult to interpret depending on how many knobs you're actually fiddling with. The preselected audio, are they from different records? To me they sound like they could be off the same album and hence mastered similarly, as another commenter seemed to imply.
the best target curve for casuals already exist. Its V shape XD
CHADrur did it again ❤
The world is right now
Can we have a server for PEQdB? I don't like the whole snobbish verification in the apple house server.
Amazing job, truly great. Haters stay mad.
Can we use your target curve?
Is it for in ear monitors because I'm not a big fan of harman in ear 2019
What about listening room curves? Are you planning to test those?
Is your shirt on inside out?
I am just waiting for PEQ sucks post
thank you for your work , sharer.
I have a question, is it the kz Vader the new king under $50?
The testing is only with music, right? Maybe people prefer different EQ when watching other content like videos, movies etc.. At least I am pretty sure I do. In my opinion music is more forgiving when it comes to tuning.
Do you plan on adding Soundcore Life Q30 to the headphone list? It was a popular budget ANC wireless OE headphone due to being able to get somewhat close to the harmann OE target with the in app 10 band EQ (oratory1990 has measurements and EQs for ANC off/on).
I actually can't import this into the Peace Equalizer on Windows. Not sure if it's a different format.
It's the correct format for peace, check if you're downloading the EQ file and not the FR file.
@@itsarqum8086 Damn, you're right. Thanks
Gee, you can import it?! lol I was manually typing everyting everytime
goated
are the iems we can select accurate? since eartips change sound too much and online databases usually swap eartips without telling us.
you can upload your own measurement if you are concerned
Can you send me the Target files?
Bro’s completed 350 trials alone 💀
not true at all actually
😂
Less bass than harman? I’d call bs
ErM iF yOu PrEfEr LeSs BaSs ThAn ThE hArMaN tArGeT yOu ArE mOsT LIkEly OvEr 50 YeArS oLd AnD fEmAle
Didn't you have your own target curve at some point? And that was more JM1 than anything?
no
No yes you did remember the Sharur target that was IE 2019 with more bass?
Needs more bass dont believe it lol
This is hugely flawed. All you're really doing here is equalizing a headphone to your preference for a single track. It's more of an preference mastering EQ target for that specific track than an overall headphone target. You need to test 5 to 10 different tracks at least and average them out like Harman did; otherwise, the equalization just sounds crap on any track other than the one you used for the A/B test.
You also need to consider the significant variations in Gras measurements. You must ensure that everyone has the same starting point for the test, but you cannot achieve that with the methods you have used. Most Gras measurements by different measurers using the exact same rig are so far off. That's something that Harman controlled for but it's impossible for you to do with this method.
Shush, you don't study at Stanford.
@@geo2160 Hahaha! If only that was a good argument. No, I'm just a mere hobbyist who's has read a lot of papers on this subject. There's a reason this hasn't been done, and it's not because others haven't already thought of it. It's because it's highly flawed. There are much better methods being worked on in audio labs around the world around measuring and simulating HRTFs. SoundID Personalization by Sonarworks did something similar, which is more accurate but still very flawed.
there are four tracks in the non upload own version..:
soundid personalization is far less accurate
Will you add t50rp mk4 to peqdb?
maybe
Is this legit?
yes peqdb.com/graphs
What about listening room curves? Are you planning to test those?
those are inherently tested