[IDrinkLava]

This is a new type of experiment I've come up with that enables people to blind test different sets of headphones on any pair of headphones as long as they EQ their headphones to the Harman OE 2018 target first. It was based on research by Dr. Sean Olive, who found that preference scores between real and virtualized headphones in blind tests were r = 0.85.


Full playlist here.

Spoiler: The methodology:

This test works by using jaakkopasanen's AutoEQ software to generate sound signatures directly from the Harman OE 2018 target. Results tend to show fairly accurate adherence up until 8kHz, at which point it reverts to a more generalized treble tilt. (Pictured below: Harman OE 2018 to Bose QuietComfort 15)
Royalty-free audio samples are then processed in Foobar2000 using the Stereo Convolver plug-in with the generated 44100Hz impulse response files split into mono files. All the samples in each video are carefully volume-matched.

Each virtualized headphone's colorations are in relation to the Harman OE 2018 target. For instance, if a particular model has a +10dB peak at 20Hz, then it will translate as a +6dB peak in the generated impulse response. The Harman OE 2018 target is used as a universal translation because of its ubiquity in various software like HeSuVi's AutoEQ function or Wavelet.

Unfortunately, I was unable to get in-ears to work the same way as there are just too many discrepancies between individual ear canal resonances and measured results.

Remember: this series is not meant to be a 100% accurate simulation of headphones. Rather, it's meant to give listeners a general idea of what different headphones may sound like.

 

[bigshot]

That is probably the only way to be able to do a blind test of headphones. Interesting.

 

[IDrinkLava]

To commemorate the announcement of the Sennheiser HD 400 Pro, this edition of The Interactive Headphones Blind Test is centered around headphones marketed as "studio headphones".

 

[bigshot]

One suggestion... Upload your videos to youtube in HD video. It assigns a higher bitrate to the audio if you do that.

 

[IDrinkLava]

Are you sure about that? When I check a 4K video in youtube-dl, the highest audio quality available is "135k opus (48000Hz)".

 

[bigshot]

I was under the impression that it includes a higher AAC soundtrack. They might have changed that.

 

[IDrinkLava]

This is gonna be the last one I share here for a while, since I've grown bored of the mindless consumerism and anti-intellectualism that dominates the audiophile hobby. I'll continue to post these on Youtube on a semi-regular basis, and hopefully the algorithm will pick up on it someday and give me more valuable input from a larger audience.

 

[KCXT]

Honest question - what's the objective here/what are you/users supposed to be getting out of this? You talk about 'more valuable input' in that last post but nowhere in this thread or in your videos are you actually stating a thesis and what the goal of the activity is.

 

[bigshot]

I suppose it's being able to get an idea of the sound of a wide range of headphone models. For me, just calibrating to Harman is my end goal. So if I've done that, I'm done.

 

[IDrinkLava]

I will be hiding my responses in spoiler tags to prevent creating any biases in any lurkers interested in this series.

Honest question - what's the objective here/what are you/users supposed to be getting out of this? You talk about 'more valuable input' in that last post but nowhere in this thread or in your videos are you actually stating a thesis and what the goal of the activity is.

Spoiler

I had two goals when I came up with this experiment. The first was to introduce the concept of blind audio testing to a wider audience with an easily digestible format [research has proven 13 minutes is the ideal Youtube video length] and an easy, universal requirement most will be able to meet. The second was to provide an alternative method for consumers to find their preferred sound without the biases that are subconsciously formed through brand names, price tags, marketing blurbs, influencers' hypetrains, common superstitions, or even just herd mentality. The three different tests serve separate purposes: the music test lets listeners find their preferences, the 3D audio test judges how differences in tuning affect listeners' HRTF localization, and the ABX test puts common myths spread around audiophile circles into practice.

I do not believe in Sean Olive's idea of a "perfect" headphone curve. In his own testing, the AKG K550 that adheres to the Harman curve about as much as the later K371 placed 4th behind the Bose QuietComfort 15, AKG K701, and Audeze LCD-2 Rev. 2. This was most likely due to a seal breach that sucked out the bass (a weakness the later K371 also shares), but it still tells me that sound preferences go deeper than just one ideal curve. From what research I have gathered from these five tests so far, the sound curve that a person tends to prefer...is actually the headphone that matches the spectral balance of whichever pair the subject owns and is accustomed to.

I will admit I have my own biases when making these tests. I tend to include one headphone that I prefer, either because I've owned it in the past or I heard its translation in AutoEQ and found it matched my tastes. Yet, despite my preconceptions of what I think the ideal sound signature should sound like, I have rarely ever seen a unanimous outcome in favor of that headphone. It is this diversity in preferences that I find valuable, which is why I seek out as many participants as possible. The input I find most valuable is from the subjects who had their biases shattered. In Music Test #1, one subject was shocked to find out how much they disliked one of the headphones after hearing a virtualized recreation of it. No doubt this subject was exposed to constant hype for the product through various influencers and forums. Another person was shocked to find that they actually liked the sound of one of the headphones after hearing so many negative things about its tuning on forums and reading its measurements.

I suppose it's being able to get an idea of the sound of a wide range of headphone models. For me, just calibrating to Harman is my end goal. So if I've done that, I'm done.

Spoiler

A very important thing about this series is that it's meant to be bipartisan among audiophile subjectivists and objectivists. I believe both sides have their own biases formed from their opposing criteria. Just the idea that the Harman target is "an averaged response of listeners' preferences", "it was initially measured from flat EQed speakers in the Harman Listening Room", and "trained and untrained listeners preferred the Harman target over the Audeze LCD-2 Rev. 2 and Sennheiser HD 800 in blind tests" is enough to make people biased in favor of it. It is an oxymoron: subjective objectivity.

In my personal experience, the Harman target leans towards a mild V-shape. It has too much SPL in the sub frequencies to be considered "flat" as well as too much SPL around 6kHz which interacts with my HRTF to create some very mild sibilance. It's subtle compared to some of the headphones in these blind tests, but still noticeable when A-Bed with my Sennheisers and Etymotics. The only objectively perfect frequency response would be one that matches the hearing curve of the individual listener, but we are still years away from having the photogrammetry that could trivialize HRTF measurements as well as fully understanding age-related hearing loss.

 

[bigshot]

I've gone over a lot of response curves and the Harman curve is very close to my ideal curve. A 1 or 2 dB cut around 3-4kHz makes it perfect.