solderdude
Grand Contributor
red pill please
-
WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
HIFIMAN Susvara Headphone Review
- Thread starter amirm
- Start date
Rhamnetin
Active Member
- Joined
- Dec 1, 2023
- Messages
- 217
- Likes
- 447
So you can't take constructive criticism... for a product you didn't design or fund? Well, that's your choice. I too happen to like a headphone that has some noteworthy performance flaws revealed by data here (and elsewhere). That just helps me understand what I like. It's irrational to get emotional and upset over it, the data is very helpful even for consumers.
How is Amir an unqualified buffoon? What qualifications does he lack? Who is qualified? How does measuring the... sound being output by a transducer, on a simulated ear rig that's far more precise than human ears, not have any bearing on sound quality?
Mr. Haelscheir
Active Member
- Joined
- Mar 16, 2023
- Messages
- 110
- Likes
- 84
I wouldn't know how the Susvara would do for a multitone distortion measurement, but here is a comparison between a substantially EQed Meze Elite with hybrid pads compared to my HiFiMan Arya Stealth and HE1000se plus an EQed lower-end Audio-Technica ATH-M50xBT:
Figure 1: REW RTA multitone signal generator settings.
Figure 1: REW RTA 4M-length rectangular window FFT noise floor with in-ear mics. The bass noise floor was a bit higher when I had the Meze Elite on my head which tends to aggravate heartbeat noise.
Figure 2: HiFiMan Arya Stealth 91 dBA equivalent pink spectrum 1/20 decade multitone distortion. The multitone distortion (the "grass" between the main tones above the noise floor) generally tracks the harmonic distortion measurements from https://www.head-fi.org/threads/hifiman-he1000-se.886228/post-17863219 (post #4,787).
Figure 3: HiFiMan Arya Stealth 94 dBA (similar to the opening of Mahler Symphony No. 5) equivalent pink spectrum 1/24 octave multitone distortion. For this simulation of a quite loud orchestral tutti (it probably reaches 100 dBA at an actual concert hall, though the cymbal takes up quite the bulk of those dynamics), the IMD floor is around 40 dB below the main signal, or around 1%, which I guess is plenty low enough at this listening level, but as you will see, isn't exceptional. Now, though my SPL meter registers louder for this denser signal with my EQed Meze Elite reference, the RTA displays a lower upper envelope to the FFT, so it's hard to say whether a denser signal or difference in crest factor actually incurs an increase in IMD percentage, else it seems to only make the IMD "noise floor" denser. Said IMD noise floor should completely block out any "details" sitting at 30 dB below, not like such would have been audible relative to the main signal in the orchestral tutti anyways.
Figure 4: HiFiMan HE1000se 91 dBA equivalent pink spectrum 1/20 decade multitone distortion. Similar deal to the Arya Stealth for similar measured harmonic distortion levels for these drivers.
Figure 5: HiFiMan HE1000se 94 dBA (similar to the opening of Mahler Symphony No. 5) equivalent pink spectrum 1/24 octave multitone distortion. Similar deal.
Figure 6: Meze Elite with hybrid pads and "V3.1 PEQ" 91 dBA equivalent pink spectrum 1/20 decade multitone distortion. IMD levels around 15 dB better than the HiFiMans despite EQing the upper midrange up a fair bit.
Figure 7: Meze Elite with hybrid pads and "V3.1 PEQ" 94 dBA (similar to the opening of Mahler Symphony No. 5) equivalent pink spectrum 1/24 octave multitone distortion. IMD levels are around 57 dB down, or 0.14%, which is perhaps properly overkill, though I wouldn't mind seeing even better results through the DCA Stealth or Expanse.
Figure 8: Audio-Technica ATH-M50xBT "V3.1 PEQ" 91 dBA equivalent pink spectrum 1/20 decade multitone distortion. This was after EQing this headphone to a very similar frequency response to my Meze Elite's EQ profile. Despite similar harmonic distortion figures below 0.1% at 98 dB above 200 Hz, this lower-end dynamic driver headphone (albeit over an unbalanced connection, whether that matters) exhibits worse multitone distortion similar in level to the HiFiMans I measured.
Figure 9: Audio-Technica ATH-M50xBT "V3.1 PEQ" 94 dBA (similar to the opening of Mahler Symphony No. 5) equivalent pink spectrum 1/24 octave multitone distortion.
Addendum:
Harmonic distortion measurements for reference, particularly the comparison between the EQed Meze Elite and Audio-TEchnica ATH-M50xBT if that reveals a utility to multitone headphone measurements.
Figure 10: HiFiMan HE1000se left driver 4M-length measurement. It so happened that my HE1000se units' left drivers had better distortion than the right independent of which in-ear mic was being used. Here, the mic used in the left ear has some bass and treble roll-off relative to the other/better mic. My Arya Stealth's right driver is a bit better than this in the upper midrange.
Figure 11: Meze Elite hybrid pads and "V3.1 PEQ" 4M-length measurement. This is improved a bit with averaging over 8 repetitions (which requires that I sit still for 3 minutes for 192 kHz sample rate measurements); the lower midrange and upper treble are noise-limited, the bass getting as low as 30 dB and the upper treble as low as 21 dB after said averaging. This headphone and EQ was used to establish the dBA calibration.
Figure 12: Audio-Technica ATH-M50xBT "V3.1 PEQ" 4M-length measurement. The noise floor line fluke is due to a noise artifact from prior to the actual test signal being played. I do not yet have an average measurement for this yet.
Attachments
Last edited:
How was the Stealth magnet in Hifiman’s flagships a quality compromise? First of all the Stealth magnet launched with the Susvara, which has a sensitivity of 83, quite a drop from the original HE1000 at 90, which you claim was superior—so that pitches your theory right out the window.
However Hifiman and Dr. Bian aren’t fools—the Stealth magnet was a rousing success when it emerged in the Susvara in 2016 (Crinacle still ranks it the second best headphone of all time, below the $60,000 HE-1!). I personally disagree with that ranking, but the Stealth magnet was so well-received that they’ve retrofitted it into every subsequent headphone in their entire lineup—the HE1000se, Arya, Ananda, HE1000, Audivina and Edition XS. Because it sucks?
The Stealth magnet arguably opened up higher frequencies above 9,000hz or so, which on earlier attempts such as the HE1000se led some to find it fatiguing—but they’ve addressed that to a great extent in their latest HE1000 Stealth, Arya Organic, Ananda Nano and Edition XS. If it was such a quality “compromise” it sure fooled all the ritzy tastemakers at Stereophile, Master’s Switch et al, because to this day, eight years past launch, they still speak in vaporous, slavish terms about the Susvara. And plenty of your apparent audiophile lesser minds are still shelling out six grand for the thing. Besides, again it has a sensitivity of 83! Which should delight you based on your hypothesis—lower than that and you’d have to power it with polonium.
What headphones do you deem worthy of serious consideration then, and paired with what amp? You must be tied down to a tank when you listen to music and be a purveyor of some rarified headphone brand that requires an atom-splitting energizer that is only reserved for gods.
As for your comments about the Utopia and Stellia, I’m completely flummoxed—the $3,000 Stellia was designed as a portable? I own the Utopia and it has a sensitivity of 104 SPL/mw, just two points lower than the Stellia at 106. It’ll play just fine off of a dongle, although of course I advocate for more robust amp pairings.
I suppose you’re going to counter this now by claiming that this sensitivity/quality issue doesn’t apply to dynamic drivers? Planars have linear impedance across all frequencies, and minimal inductive resistance—liberating the diaphragm from the voice coil theoretically lends them to more power efficiency by design. But before your posts I had been completely unaware that in 2024 headphone technology was still too primitive to produce a quality headphone that didn’t require extremes of power! The audiophile market has expanded so exponentially in the past decade to wider markets that this has demanded that the technology advance to accommodate more accessible amplifiers. Do you only own electrostatics?
Based on your comments en masse I question whether you’ve even heard all these crappy, efficient headphones you’re denigrating. Your profile indicates that you’ve been a member here since December of 2022. What have you been reading?
Last edited:
Mr. Haelscheir
Active Member
- Joined
- Mar 16, 2023
- Messages
- 110
- Likes
- 84
An important amendment regarding my in-ear speaker HRTF measurements. I had lately used threshold of hearing EQ whence I had found that my blocked ear canal measurements were indeed failing to account for differences in canal gain between the 30-degree speaker incidence and the 90-degree headphone incidence and coupling. As such, when using in-ear mics to calibrate headphones to a speaker response, though the blocked canal measurements may show greater ear gain levels than measured through existing neutral headphones, after compensation, the ear gain levels will become more comparable. See "Calibration using threshold of hearing curves" in https://www.head-fi.org/threads/rec...-virtualization.890719/page-121#post-18027627 (post #1,812).
You cannot disprove a person's preference for a sound that objectively measures worse (lower fidelity, that is) with objective facts. Preference and taste are fundamentally subjective. I guess that is the point.
If somebody just prefer the horrendous airline free earbuds to everything else, what measurements can you use to convince that they are wrong? Individual preference is just radically subjective and no objective fact can or is needed to prove or disprove it. Although "average preference" could be objective and that objective correlations might be established between measurements and average preference. However, this is not done here. Harman investigates the correlation between frequency response and average preference. But I don't think anyone has done enough scientific research on the correlation between group delay or harmonic distortion at 114db and average preference. And also, there is no reason that an individual buyer should be much concerned with the average preference in the first place, unless their preferences match the population average. This is why interpreting the measurements here (or anywhere) as objective and absolute indication of sound quality that is relatable to yourself is misguided.
Nobody will buy these before hearing them in person. But if you have heard them, and you have enough money to spend, and indeed you subjectively prefer this to everything else that is cheaper (though this part better be verified through double blind testing), there is nothing wrong with buying it, and nothing wrong with the company capitalizing from these situations.
It means that some of the things measured here are not that important sometimes for personal preference. For example, distortion at 114db is hardly relevant for anyone who doesn't want to be deaf. Distortion at 60-80db probably much more relevant for most people, but the vast majority of headphones won't perform bad at these levels, so there is not enough differentiation. Messy group delay, as Amir has even observed, sometimes correlate with great spacial capabilities. Also the harman target is supposed to predict "average population preference frequency response", but your own preference target could well be something different.
Amir's reviews consistently omit discussion of treble above 8khz. He himself acknowledges that the measurement is inaccurate and that his hearing is unreliable in that region. However, I think most of the extra stuff that more expensive headphones can offer on top of the good budget options resides in the treble. The two zeros don't have good treble compared to well received iems in the $200-500 price bracket, for example. There are plenty of iems that have very similar response to those, with the difference lying almost entirely in the treble.
Being fast has more to do with the frequency response; transients are after all fourier transformed into pure sine waves by the inner ear, and fast transients require the correct amount of treble frequencies to realize. but spatial properties do correlate with group delay, this even Amir has observed.
If you listen at any health volume, no peak ever is going to reach 114db. That is absolutely one of the most irrelevant for real use cases for most people in amir's measurements.
I think only the sr009 and l300 can be described has having a 1k peak. Lambda pro doesn't have it. The sr007 is better described as a large dip between 2k-4k.The reason Stax headphones are found to have 'faster transient response' is the peak nearly all of them exhibit at around 1kHz.
Add the same peak to a HD600 and it sounds 'faster' or lower that peak on the Stax and the 'fastness' of the Staxes is removed.
Arguably the HD600 is 'fast enough' as it can do well over 20kHz. It is a long standing myth about Stax headphones (the speed and transients).
The 'attack' people hear is in the 1-3kHz range. No 'fast membrane' needed for that.
Below some Stax headphones (and HD600/650) to illustrate the point:
and the HD600:
HD650:
solderdude
Grand Contributor
Lambda pro has a peak around 1.5kHz, same effect.
007 has it too but also has a dip in the earcanal gain band so it is less obvious as it lacks in clarity (which some folks prefer).
007 has it too but also has a dip in the earcanal gain band so it is less obvious as it lacks in clarity (which some folks prefer).
We're not necessarily talking about peaks reaching 114dB in order for the 114dB distortion line to be relevant in Amir's measurement. It's more about bass EQ. Let's say absolute worst case scenario you have to put in a +20dB bass EQ if you've got a pair of headphones that rolls off in the bass a lot when combined with someone who is a major bass head that likes even more bass than Harman Curve. In this scenario your 1kHz point (for arguments sake) would in the region of -20dB down due to the bass EQ (negative preamp), so 114dB minus 20dB equals 94dB at 1kHz. So in that example Amir's 94dB blue line would be relevant for the mids & treble, and the 114dB green line would be relevant for the bass, and that would be worst case distortion because music is not recorded at 0dBFS. So the 114dB green line in Amir's distortion measurements are very rarely relevant, but in outlier cases of headphones that roll off a lot in the bass combined with someone who likes above Harman bass levels then 114dB green distortion line could be relevant, but rarely. The example I gave in the post of mine you quoted which is my own use case then Amir's blue 94dB is only the real distortion measurement of relevance to me because I don't boost bass beyond Harman and I don't listen at loud levels. Me personally I pay attention to Amir's blue 94dB line and the red 104dB line (to allow for "nice to have" overhead).
Last edited:
solderdude
Grand Contributor
Indeed but have a look at this:
This part of the recording below was playing uncomfortably loud. It is the type of level where you turn up the volume because you either like the song very much or want to listen ‘deep’ in the recording to evaluate.
Not a level you would gladly endure for the whole song though.
This resulted in the following (rounded off) numbers:
RMS levels = 510mV = 0.81mW = 97dB (average)
Peak levels = 9.16Vpp = 3.24VRMS = 33mW = 113dB
calculated DR = 16dB for that 10 sec part of the song
(this was the intro of the song by the way)
And as most of those peaks are low frequency low (bass) and this is without bass boost you might be able to see why testing (frequencies below 200Hz) as being relevant.
Of course, no one will listen at 114dB average let alone 114dBA.
of course in recordings there can be 114db peaks. The question is will anyone play them back so that the peaks measure 114db at the eardrum?Indeed but have a look at this:
This part of the recording below was playing uncomfortably loud. It is the type of level where you turn up the volume because you either like the song very much or want to listen ‘deep’ in the recording to evaluate.
Not a level you would gladly endure for the whole song though.
This resulted in the following (rounded off) numbers:
RMS levels = 510mV = 0.81mW = 97dB (average)
Peak levels = 9.16Vpp = 3.24VRMS = 33mW = 113dB
calculated DR = 16dB for that 10 sec part of the song
(this was the intro of the song by the way)
And as most of those peaks are low frequency low (bass) and this is without bass boost you might be able to see why testing (frequencies below 200Hz) as being relevant.
Of course, no one will listen at 114dB average let alone 114dBA.
solderdude
Grand Contributor
114dB SPL in the bass... yes, when shortly playing a song at loud volume and useful to measure. If only to check useful limits of the headphone in question.
Many percent distortion in the low bass is not detrimental but one has to realize that 'riding along' on those excursions there is also the rest of the spectrum which (at that moment) also is modulated a bit. Fortunately those peaks are very short and loud and the rest of it (just after the peaks) is at much lower levels.
Distortion is not a big issue for planars but for some dynamics it is.
I have measured (observed) compression (illustrated by increased 3rd harm distortion) above 80dB SPL already in the bass. This means that even at 70dB average levels there is already an increased amount of distortion. The audibility of that remains to be seen because of the short moment events this is happening in.
The measurements I posted are real and show this SPL can occur IRL. I even described how loud it felt. Of course not all energy was in the bass and some of it was also in another spectrum but the peak SPL reached (the excursion needed) shows measuring a headphone at 114dB SPL is a good idea. Granted, one could stop the measurement above 300Hz.
Many percent distortion in the low bass is not detrimental but one has to realize that 'riding along' on those excursions there is also the rest of the spectrum which (at that moment) also is modulated a bit. Fortunately those peaks are very short and loud and the rest of it (just after the peaks) is at much lower levels.
Distortion is not a big issue for planars but for some dynamics it is.
I have measured (observed) compression (illustrated by increased 3rd harm distortion) above 80dB SPL already in the bass. This means that even at 70dB average levels there is already an increased amount of distortion. The audibility of that remains to be seen because of the short moment events this is happening in.
The measurements I posted are real and show this SPL can occur IRL. I even described how loud it felt. Of course not all energy was in the bass and some of it was also in another spectrum but the peak SPL reached (the excursion needed) shows measuring a headphone at 114dB SPL is a good idea. Granted, one could stop the measurement above 300Hz.
They of course "occur in real life". I could make it even louder and subject my ears to it. But I won't do that, this is clearly not a healthy (or for that matter, enjoyable) level to listen to music. I don't think it should be considered as part of regular usage of such a product. I don't think we need to skew our assessment of headphones so heavily in favor of performance in these very rare scenarios that don't last long.
solderdude
Grand Contributor
The thing with testing is that you want to know how it performs in 'normal' circumstances AND near the extremes where is (sometimes) might be used by some people.
How everyone (mis)uses their headphones and ears is an entirely different matter and is not related to measuring certain aspects at the extremes.
How everyone (mis)uses their headphones and ears is an entirely different matter and is not related to measuring certain aspects at the extremes.
Similar threads
- Poll
