CHANGSTAR: Audiophile Headphone Reviews and Early 90s Style BBS
Lobby => DIY => Topic started by: ihasmario on October 12, 2013, 02:44:51 PM
-
Hi Gang,
I've been working on building a binaural dummy head for recording purposes, figured it'd be a fun thing to post here. At the end of this post, you'll find an mp3 recorded showing the progress of the project and how pronounced the effect is so far. I'll also detail the costings for those who are interested in this kind of thing.
Prior to this I had a Schneider disc, which I will post a similar test of later when I am happy with the head.
At the moment the head is bare bones - literally a hollow dummy head with small electret capsules in the ears. Here are some pictures.
(http://i39.tinypic.com/2hgu4xv.png)(http://i43.tinypic.com/16c5ncg.png)
(http://i41.tinypic.com/r2699d.png)
Here's an mp3 - it includes a click test, guitar recorded several ways (badly - bad playing and one recording session I was too close. Three recordings at the sound hole, one at about fret 12), and then a song recorded to demonstrate the hollow echo a bit more closely, which is currently the major flaw. The song is aiming to be a perfect reproduction of the original, not a binaural version. I'll do some more sophisticated tests later.
Here's the mp3
https://app.box.com/s/qm0finguto7lpwr2ewq2 (https://app.box.com/s/qm0finguto7lpwr2ewq2)
If you are interested in comparing the song to the original, it's available for free from the record companies site
http://www.higherlivingrecords.com/index.php/releases?id=75:elin (http://www.higherlivingrecords.com/index.php/releases?id=75:elin)
What I'd like some people to do if they have time, is to try and identify what motion the clicks are doing in the first section. I will be posting the answers in about a week if enough people reply. The aim is to determine without bias, how well the head is producing an image at this stage.
- The first set of clicks begins at 0:00
- The second set begins at 0:11
- The third set begins at 0:20
- The fourth set begins at 0:26.5
- The fifth set begins at 0:32
Thanks everyone
Parts list
Dummy Head - dis-m121 - Bought for $20
Microphones - Naiant U-M 8mm omnidirectional capsules - Cost $15 each
Preamp - LineAudio MP8
Interface - Audiofire 12
Total head cost so far: $50
PS I started to teach myself guitar very recently so the playing is very sloppy.
:)
-
Space reserved for the following head moficiations:
- Filling in the head to remove hollowness
- Implementing proper ear canals
- Sculpting better ears (more pronounced): Possibly removal of the ears to include silicone ears
- Sculpting of the face to accentuate front image.
- Adding hair or windshields to the ears for pops that occur close to the ear.
- Attaching to stand, a better set of recordings at a greater distance.
:spank:
-
Space reserved for new recordings and comparisons to schneider disc setup.
In the meantime, he's a sample recording of the same capsules applied with a Schneider disc. Most of this isn't head on though. Obviously the guitar had better strings for strumming on back then (Elixir PBs).Featured in this recording is; Guitar (strum), Guitar (hybrid), Guitar (Finger pick), Tenor Saxophone, Cornet. Mostly recorded front-on and one shout so excuse sloppiness been a bit busy to maintain/develop them all.
Here's the temp mp3
https://app.box.com/s/gz8u1cb6rx8s7f98et6u
Bonus points if anyone can guess the group whose songs feature heavily :)
You can tell there is an absence of a hollow echo that the head current has.
Thanks
-
Interesting project! The head is a bit androgynous... :-Z Not that it matters.
-
Thanks! I'll check the tunes tonite! Keep up the great work!
-
Awesome project! Keep us posted!
Oh, and sorry 'bout the new avatar... I couldn't resist.
-
Oh, and sorry 'bout the new avatar... I couldn't resist.
This one is better popcorn
Any ideas on what I should fill the head with? I have a spare sound damping foam that I can feed in there, but I wonder if stryofoam balls might be a cheaper and possibly better option (also easier to remove if I don't like it), as it would add more dead air space than open cell foam.
-
Very nice and thank you for providing the links for the resources to build this.
-
Listened using equalized Hifiman RE-400 IEM with custom "tips".
Without crossfeed:
The three left-to-right motions are nearly indistinguishable, there's no depth and almost no height to them. If I were to guess, first was front, second above, third behind the head, but all three sounded wrong, passing through head instead of around.
If I'm right the first two were circle-like outwards from the left/right ear, the depth is hard to describe, but sounds like 10-15 cm at most. Music samples of the first music are (counting from the ear plane) 90 degree - in front, 30 degrees tp the right, 30 degrees to the left, 60 degrees to the left? and again front. Cannot say anything about depth/distance. All samples sound in-head and very flat, much more than the original mix.
Original mix sounds a bit "widened stereo", but not in-head at all.
With crossfeed:
The original music sample sounds properly forward focused and close to the non-crossfed studio version in terms of depth retrieval. Much more centered.
Suprisingly, adding crossfeed to the head samples makes it sound kind of correct, first and second click being moving back and towards the ear about 1.5m (starting from the ear), third being in front close to the head - sounds very discontiguous, with a hole in the middle; fourth being in front further away?, less of a hole in the center; fifth being behind the head, sounds the most correct to me, almost no hole; The guitars are like sitting at a distance of ~1m in front, about 1m to the right; 1m to the left but closer, bassier?; right below the face or at it, in front.
Music also sounds much better, heck, I'd say better in some respects than the original sample via crossfeed - natural room-like reverberation - though it sounded like a bit of a V-shaped bass cut, too nasal with a hole in the middle.
All in all, I think the recordings do sound quite clean, so the capsules and recording environment are reasonably good. It is a very quiet recording though, needs a bunch of dB boost.
This would suggest that the head is highly ineffective in capturing crossfeed - the microphones not being omnidirectional enough, having a half-sphere directionality, not really omni? That or the head is not sonically permeable enough or is internally too reflective.
HRTF must be more or less correct, since with crossfeed front/back is fine I think. Height is uncertain, I'm not sure you've tested it at all.
-
HRTF must be more or less correct, since with crossfeed front/back is fine I think. Height is uncertain, I'm not sure you've tested it at all.
Thanks for the large and detailed feedback, will PM you the results if you like.
RE: Testing. There hasn't been any testing done so far, there are a few very obvious problems (the head seems to be resonating) that I need to deal with before any of that. The capsules also have a slight upward tilt.
RE: Imaging problems, I expected it wouldn't be as clear as I hear it (since I know which is which...). I think it has to do with the depth of the mics (given that the head has no canals at the moment), which I think would translate to less treble being lost in any direction. Going to the hardware store on Tuesday to buy some rubber tube. Haven't tested the mics atm either, will probably test them later in the week with the Schneider disc, comparing front and back. :)p4
EDIT:: Just listened to the clicks with my ER4S and found similar problems to you that weren't obvious on headphones. The direction of the clicks also changed compared to what I knew they were, most were projected upward and slightly forward, but like you say mostly through the head rather than slightly around.
EDIT::: A few other things I havent done/request for people listening, try apply a generic diffuse field curve - I haven't taken one of the head yet. I am still learning about binaural recording TBH so it's just dipping my feeti n
-
DF is the wrong curve in this case - there is a diffuse characteristic from the room. What should be applied instead is independent-of-direction (ID) curve or similar ear canal simulation. That'd go like this: boost 3k to 6k range about 4.5 dB, (less than DF), do not cut highest end or bass. Alternatively, for IEMs, wear a deep insert IEM shallow but still sealing - way simpler. (Except RE-400, which seems to be designed for deep fit but becomes linear in shallower fit. Works with Brainwavz B2.)
It helps a bit, but the actual issue has something to do with phase/group delay, head shadowing and not magnitude - that's why crossfeed helps so much, especially if my crossfed positioning estimates are mostly correct. (PM me please with the right answer, I won't tell anyone. p:/ )
-
DF is the wrong curve in this case - there is a diffuse characteristic from the room. What should be applied instead is independent-of-direction (ID) curve or similar ear canal simulation. (boost 3k and 6k, less than DF, do not cut highest end) Alternatively, for IEMs, wear a deep insert IEM shallow but still sealing.
It helps a bit, but the actual issue has something to do with phase/group delay, head shadowing and not magnitude - that's why crossfeed helps so much, especially if my crossfed positioning estimates are mostly correct.
Cool thanks for the info. I'll do some more sophisticated recordings (especially from a greater distance, since all of these were extremely close) to see how much of an image I can get after I have added canals.
I can't really figure out how to mic an acoustic from a distance (especially fingerpicking), so it'll probably be vocals or sax. :)p4
-
This one is better popcorn
I disagree. The one I gave you had a creepier stare, because the head was tilted sideways slightly and more of the bald head was visible. It had more of a Max Headroom glitched out quality to it.
-
DF is the wrong curve in this case - there is a diffuse characteristic from the room. What should be applied instead is independent-of-direction (ID) curve or similar ear canal simulation. (boost 3k and 6k, less than DF, do not cut highest end) Alternatively, for IEMs, wear a deep insert IEM shallow but still sealing.
It helps a bit, but the actual issue has something to do with phase/group delay, head shadowing and not magnitude - that's why crossfeed helps so much, especially if my crossfed positioning estimates are mostly correct.
Cool thanks for the info. I'll do some more sophisticated recordings (especially from a greater distance, since all of these were extremely close) to see how much of an image I can get after I have added canals.
I can't really figure out how to mic an acoustic from a distance (especially fingerpicking), so it'll probably be vocals or sax. :)p4
It depends what microphone type you use, the room, and what the end result will be.
If you are using your omni's, get someone whose hearing you trust to walk around the room and find a spot where it sounds great. Place omni mic there. Done. :)
-
It depends what microphone type you use, the room, and what the end result will be.
If you are using your omni's, get someone whose hearing you trust to walk around the room and find a spot where it sounds great. Place omni mic there. Done. :)
If I am using the tiny pencil-style omnis, how will I control for self-noise being amplified with a quieter source like a guitar from a distance? I have some 10mms which would have less self noise... I'll try and post a distance strum recording (not that the strings are good for strumming...) tonight to test the image.
-
It depends what microphone type you use, the room, and what the end result will be.
If you are using your omni's, get someone whose hearing you trust to walk around the room and find a spot where it sounds great. Place omni mic there. Done. :)
If I am using the tiny pencil-style omnis, how will I control for self-noise being amplified with a quieter source like a guitar from a distance? I have some 10mms which would have less self noise... I'll try and post a distance strum recording (not that the strings are good for strumming...) tonight to test the image.
The easiest would be to do a multi-mic set-up and multi-track it. Then do a blend of the ambient mic with the direct sound mic until you have something you like.
This is where mic quality really matters and that's why the better omni's cost $10,000 or more - they have virtually no self-noise.
-
True, the really good omnis are expensive, especially measurement grade ones.
The best cheap ones I know of go to 70 dB SNR (Dayton EMM-6) which would be about ~30 dB (A) self noise at calibration, which is not astonishing - like a very good cassette. No silly things like mains hum there.
But if you toss in another order of magnitude, $650, then you can get an Earthworks M30, which has 22 dB (A) self noise, should be better than almost all rooms and is a bit more sensitive than EMM-6.
And it gets a bit better still higher up, but frankly I wouldn't break the bank too badly on the omnis.
-
Lack of updates due to being busy. I went and bought some silicon tubing to create canals, it resonates horribly and sounds worse for a small increase in image.
Looking for alternatives or ways to dampen the canals. May end up binning the project for a while due to employment (or lack thereof)
-
Silicone nozzles must be longer than you'd expect - the ear canal volume has to match and that's 0.7-1.3 cm^3. (Not 2cc, that's way too large.)
Also watch out for edge effects, do flare these in a horn pattern, and watch out for the seal with the capsule.
I used 4mm diameter tubing and attached rubber "horn" with good results for one EMM-6. Without the horn, you'll get notable 6kHz half-wave resonance or null.
-
Silicone nozzles must be longer than you'd expect - the ear canal volume has to match and that's 0.7-1.3 cm^3. (Not 2cc, that's way too large.)
Also watch out for edge effects, do flare these in a horn pattern, and watch out for the seal with the capsule.
I used 4mm diameter tubing and attached rubber "horn" with good results for one EMM-6. Without the horn, you'll get notable 6kHz half-wave resonance or null.
Which end does the horn go entrance or exit? Do you remember where your data is from so I can read it?
-
Uh, the data was a cheapo measurement of a headphone (my HE-500) without a real ear, made using REW. Pure silicone tube worked decently well for IEMs (except 6k dip), but terribly for headphones. I used a bit of plywood for seal dampened with some foam, not ideal I know, real silicone ear will come later. However, after adding a horn it worked pretty decently, though not perfect and way better for IEMs. I've reduced ear tube length to take the "outer ear canal" volume into consideration.
The horn is wider edge at the "ear" like an ear canal - not a reverse horn.
1.5 cm far end diameter, 3mm near end diameter, 2.5 cm long, otological for looking into ear canals, plastic (PET?). Costs a dollar or two.
I will make a real copy of my ears sooner or later.
-
Random idea for this sort of thing:
- Make cast of head - http://www.smooth-on.com/gallery.php?galleryid=117 (http://www.smooth-on.com/gallery.php?galleryid=117)
- Fill the cast with ballistic gel a la mythbusters - http://how2dostuff.blogspot.com/2006/01/how-to-make-ballistics-gel.html (http://how2dostuff.blogspot.com/2006/01/how-to-make-ballistics-gel.html) (Also make sure to include wiring harness for later
- Seal a pair of custom iems in cellophane and use their shape to cut an approximation of the ear canal into the dummy
- Wrap the mold in artificial skin made from cow collagen and shark cartillage - http://www.ilstraining.com/bmwd/bmwd/bmwd_it_00.html (http://www.ilstraining.com/bmwd/bmwd/bmwd_it_00.html) The dressing should adhere and allow for a measure of elasticity. (Also now this thing is bound to look creepy as hell. Might as well add a wig and an artificial eye or two)
- Perform an operation to add the two mics at the end of the ear canal, connect them to the harness
- Find a mannequin, decapitate it and recapitate with the dummy measurement head
- Give this abomination a name. Then hug it. Hug it tightly, for it is your new HRTF clone.
- ...
- Profit.
-
Way ahead of you. Made a simple crummy model, because I needed it yesterday.
See this post: http://www.changstar.com/index.php/topic,1028.msg32777.html#msg32777
It's an attachment to a speaker made to actually emit sound, not receive it, but a similar method could be used to make a synthetic head. The size is quite a big one, but still possible. I've seen people with slightly bigger heads still.
The rags are actually the dampening. They work extremely well at low frequencies, so the resulting mouth stream is highpassed, counteracting the proximity effect somewhat (but not completely). It can be linearly compensated reasonably easily, mostly there is too long resonance still due to a cheap speaker, but REW can bring it to near perfection with just 6 peaking eqs, mostly trimming midbass and bumping 6k up.
Reasonably low nonlinear distortion due to this - measured with REW's RTA function and ECM8000 as around 0.1% D2 over the Behringer C50A speaker's baseline, which isn't that great either. Polar characteristic is very similar to real mouth, I haven't measured that though, just gauged it with ears. No anechoic chamber for that measurement.
I'll make a better version for a true HATS dummy lookalike, except prettier. :&
As amusing as dousing someone in silicone is, that is not necessary.
Cellophane is also not necessary - use a demoulding compound, water with soap or watered down wax, just don't forget to seal the air tubes with e.g. scotch tape.
Simpler way:
- Get a gypsum or plastic skull. You can use cosmetic plastic cheekbones instead.
- Use two bowls to cast lower and upper half of the mouth as well as back of the head in silicone.
- Put the skull in.
- Fill the necessary parts with PU foam or ballistic gel if you feel like spending more.
- Cut out (or fill before putting PU/gel in) nasopharyngeal (nasal) cavity.
- Attach some nose
- Mould some ears from polyclay (sulfurless is recommended)
- Put in fake or real ear canals with the right volume, such as 1.2cc
- Add fake silicone lips. Name it Jenny.
- Keep on rockin'! headbang
Have a peek of our measurement "setup" for taking acoustic echo samples. Please excuse the crude method of attaching it to the speaker.
Note: this is confidential, don't get those nasty buggers lawyers on my ass.
The output was compensated with REW (generating impulse response) and SIR v1 was used to apply the filtering to the mouth signal.
The room is not anechoic to simulate real conditions. The other end wasn't pictured by me, but it lacks the head as it's not as critical there to have correct polar and reflection characteristics.