Frequently Asked Questions
Feel free to send us your questions.
Q: We humans can only hear out to 20 KHz, and only then if you are about 6 years old... So why use 96 KHz sampling?
A: This statement is generally true enough, but it has nothing to do with our choice of 96 KHz processing. By using a high sample rate, we increasingly oversample the lower frequencies of interest. This leads to smoother recognition of sharp-attack sounds, like drum-strikes. It also permits more accurate digital filtering at lower frequencies.
The difference in quality of restoration when using 48 KHz sampling, compared to 96 KHz, is very striking. At 48 KHz, the sharpest attack sounds are "in-your-face". But with 96 KHz processing these same sounds are delicate, exhibit subtle detail, and are still plenty loud enough.
Q: You state that unwanted harmonic distortion products are below -130 dBFS. Isn't that dependent on the amount of spectral "bending" that you do?
A: Yes, indeed. This figure of merit was measured under conditions corresponding to corrections for a hearing threshold elevation of 50 dB at 6 KHz. That represents a pretty serious hearing loss at that frequency.
The test tone was a 1 KHz -14 dBFS Sinewave (or, 0 dBVU on the K-14 system). And it is true that even under these strong corrective conditions, the 1 KHz signal would be receiving less "bending" than a 6 KHz signal. But that is true in general. Sensio-neural hearing loss generally rises, sometimes quite strongly, at higher frequencies.
Q: Why are you using outboard equalizers when you could be doing it all in the digital domain?
A: Our reasons for this are several-fold:
1. Audio professionals rave about the quality of the Massenburg 8200, and even moreso about its mastering big-brother, the model 9500.
2. While it is true that you could do a reasonable job of providing EQ, especially at 96 KHz sample rate, there are still digital quirks to contend with, compared to a fine analog equalizer. E.g., throughput latency. That may interfere with live performance requirements. A fine digital equalizer like our 10-band unit is also power hungry and it cannot run at highest quality levels while sharing the CPU with System 5000.
3. Under some conditions of strong hearing corrections, the amount of boost required at higher frequencies can rise to as much as 24 dB under low level signal conditions. We have two choices in this case:
(a) We can attempt to provide all of the reconstructive processing in the computer. But if we do this, then we face the possibility of strong clipping unless we reduce the internal processing levels by as much as 4 bits. There are other sources of dynamic range erosion in our computations that further increase this by another 3-4 bits. Hence we would be left with a signal equal to 16-bit quality on output.
(b) If we de-emphasize the spectrum in the computer, and then rely on an outboard analog equalizer to re-emphasize by an identical amount, then we can avoid the 4-bit erosion due to high gain levels in the corrections.
The output signal will be very similar to the incoming signal, or perhaps just a bit more "white", and so we have effectively a 20-bit audio signal. (Remember the other effects that erode by another 3-4 bits?). A 20-bit audio signal gives you a pristine dynamic range of 120 dB, which is more than our ears can accommodate, within safe listening levels.
Clearly in these circumstances, using digital equalization would gain nothing. It still has the same headroom limitations as our corrective processing. So a high quality outboard equalizer, like that second Massenburg 8200 inserted between the Muse Receptor and the headphone amplifier provides additional headroom for us in the analog domain.
Q: 500 MFLOPS!? What's the big deal here? Why don't we just insert a Graphic Equalizer between the sound and our headphones or speakers?
A: WHOA!! You mustn't attempt this! It is exceedingly dangerous.
Damaged hearing acts like a non-linear gain expander. This is known as hearing "recruitment". (See the diagram below) As such, any direct equalization will only be heard correctly at one particular loudness level. As sounds grow softer, they become too soft. As the volume increases, the sounds become too loud.
Our system compensates precisely for the recruitment and keeps sounds in correct proportion to the listener. We implement ear safety protocols to ensure that no sounds reaching your ears will ever approach damaging levels. In fact, when our system is properly calibrated by us, you can expect never to hear any sustained levels above 90 dBSPL. And these levels will only be reached during loud crescendo passages in the music. In general, as long as a K-meter stays below the red-zone, you are completely safe.
Using a simple equalizer for hearing compensation is foolish because it applies constant boost, no matter how loud the signal is. If you were to need 20 dB gain at some frequency and loudness level, then during those crescendo passages you will be exposing your ears to 110 dBSPL. That may be not only painful, but severely damaging after a rather short cumulative exposure.
Please! Never attempt to correct your hearing with a simple equalizer.
Diagram illustrating the nonlinear
gain expansion of hearing recruitment. The black reference line represents perfect hearing. The
orange curve is the recruitment for a threshold elevation of 30 dB, while the red line shows
that for a threshold elevation of 50 dB.
In both cases, notice that as we approach the loudest sound levels, the hearing sensitivity
approaches normal levels. But that doesn't mean that by increasing the volume you solve all problems...
Remember that the amount of recruitment present in your hearing varies with frequency. You probably need very little help in hearing the bass, but you need more to hear the highest treble. Raising the volume will blast you out of your seat with the bass before you get to hear that treble range.
Here is a graph showing
exactly what happens when you either raise the volume level, or utilize static equalization
in that frequency band exhibiting, say, 50 dB of threshold elevation. Here we are showing a
boost of 20 dB.
Notice that there is only one input loudness level at which the perceived loudness is
correct. That's where the red curve crosses the black perfect hearing line. As sounds grow
louder, they are perceived as much too loud. And as sounds become softer, they rapidly
disappear below your threshold for perception. So what have you really gained by this? Not much!
This also illustrates why using simple equalization, even if you could keep it safe for your ears, will produce an incorrect mix at your console. And since this occurs in varying amounts at different frequencies, you can't hope to achieve a correct mix in this way. It is utterly futile... unless you have a System 5000.
The System 5000 precisely rectifies the recruitment curves (gain expansion curves) in each of 100 overlapping quarter-Bark frequency bands, so that the perceived result matches those black reference lines in the graphs shown here.
Q: What are those odd looking dB units on your graphs? dBHL and dBSL?
A: These are decibel units that refer to Hearing Level (HL) and Sensation Level (SL). When audiologists measure human hearing they utilize a dBHL scale that is offset by 19.5 dB from the absolute zero of human hearing thresholds. If we call that absolute threshold 0 dBSPL (Sound Pressure Level), at 1 KHz, then a 0 dBHL is actually 19.5 dBSPL.
Why do this? Well normally we live in a world that never permits you to exercise your hearing at the absolute threshold levels. To do that you must live in an isolation booth. Our ambient surrounds are realistically much higher intensity levels. Human whispers at 3 feet approximate 20 dBSPL, or 0 dBHL. If your hearing is within 20 dB of 0 dBHL, then for all practical purposes you have normal hearing.
The dBSL scale refers to how the sound is sensed by a person. A person with perfect hearing, on being presented with a tone at 30 dBHL, will respond that they hear it as it is, and hence they will sense a 30 dBSL intensity. People with recruitment can't hear anything until you exceed their threshold elevations, and then the sound intensity appears to rapidly increase with very slight actual increases.
So a person with a 50 dB threshold elevation, doesn't hear anything until the sound intensity levels exceed 50 dBHL. At 55 dBHL, they may report hearing a sound that corresponds to 35 dBSL. In other words for that small 5 dB increment, their hearing has covered about 35 dB of territory.
Q: What are some of the elements of the 5000 System process? Is it simply a matter of rectifying the recruitment curves?
A: It is primarily a matter of rectifying these recruitment curves, but we also take into account the frequency dependent nature of these curves. The famous Fletcher-Munson curves so often cited can serve as a basis for this frequency dependency. (We actually use a somewhat more sophisticated model.)
The Fletcher-Munson curves are often illustrated as having essentially the same shape at all sound intensity levels until you reach the pain threshold at 120 dBSPL. Well, as they say at the carnival, "close... but no cigar..."
Our research has shown that there is a gradual flattening of these curves as you move from threshold levels toward high-intensity sound levels. That's one refinement of our models.
Another takes into account the "Masker Spreading Function", which describes how sound in one of your critical frequency bands also affects nearby critical frequency bands.
Our models for human hearing show us that the recruitment curves have a rather complex shape that depends on the measured threshold elevation. As you can see from the graph above, they become ever steeper as that threshold elevation climbs. But more than that, the characteristic exponents of these curves change as well.
And let's not forget the system dynamics. This is at least as important as rectifying recruitment curves. The way our hearing responds to abrupt changes must be taken into consideration. Also, we have to be very mindful of the generation of IMD products as a result of rapid gain changes. So this is an equally complex part of our system.
Our research over these past six years has been very involved and detailed. Some of our researchers have backgrounds in radio electronics and they found that radio figures of merit, such as IMD Intercepts, can also be usefully applied to our own "receivers". IMD measurements have provided some of the experimental evidence supporting our theoretical hearing models. It has been a very involved and complex journey.
So, yes, there is much more to it than merely rectifying these recruitment curves.
Q: Why don't you release a VST plugin that can run on a standard computer DAW?
A: We have!
This system is especially adapted for use on personal computers. It processes the most common sample rate signals used by CD's and internet sound files (44.1 kHz).
However, it has always been our aim to provide extremely low processing throughput latencies, so that these hearing systems could be used by live performers and professional audio engineers in the mixing and editing of soundtracks.
Attempting to run our high-end (e.g., System 5000) code on a conventional computer means that we are subject to the extraneous processing overhead of the computer's operating system. The Muse Receptor, and the Soundart Chameleon, represent carefully designed, dedicated audio processors. As such they can offer the extremely low throughput latencies sought by us. Conventional computers cannot, or at least, they cannot do so reliably.
Then too, if we run on an individual's computer DAW, then we are unable to control the contention for processor cycles as the user loads up on lots of other plug-ins and many editing tracks. The support logistics become nighmarish. Our algorithms, even the simplest ones, require substantial processing power, and typically require a 50% processor load on even the fastest computer workstations.
To make matters worse, a typical Mac G5 (a preferred audio DAW computer) runs these same algorithms at around 80% loading -- making it virtually impossible to perform any other useful work on the computer, and causing substantial dropouts, clicks, and stutters, depending on whatever else the computer needs to do.
Processors heat up substantially when you run them this hard. Our fastest Pentium based workstation in the lab here, kicks in with a very harsh sounding fan, and will overheat if this kind of performance is required for any length of time beyond just a few minutes. In contrast, our Muse Receptors and Sound-Art Chameleons typically run our algorithm code for days and weeks on end without incident.
Finally, even if you could overcome these problems with conventional computers, doing the hearing restoration on them means that you cannot properly apply any outboard effects processing, post DAW. The hearing restoration needs to be applied at the very end of the audio chain, just ahead of the headphones and speakers.
So, having a dedicated high-performance audio processor on the outboard is really the only way to go.
Q: What is V-Tuning?
A: V-Tuning is our proprietary system for tuning up a hearing restoration system for persons with typical symmetric sensioneuro hearing imperfections.
We had some astonishing breakthroughs during our laboratory research into the nature of human loudness perception and sensioneuro hearing impairment, as a function of frequency. We cannot describe in much more detail as this exciting development is under strict internal security until pending International Patents have been awarded.
Suffice it to say that the proper tuning of a system with stereo 100-bands typically requires the correct setting of perhaps as many as 1400-1600 different parameters. Or if you want to view the problem from the perspective of available audiology measurements, then approximately 200-300 dependent variables.
Either way, our discoveries allowed us to orthogonalize these solutions so that in effect we have a single parameter family of correction equations across all the 100 bands in each stereo channel. This is an absolutely remarkable simplification. It permits the end user to control his/her corrections using only a single knob adjustment, and then simultaneously and correctly setting all 200 frequency bands of correction for their hearing characteristics.
Now there are sometimes peculiarities in some listener's hearing that may make this form of simplified correction unsuitable. For those listener's we revert back to setting either 12 (symmetric hearing) or 24 (asymmetric hearing) threshold elevations at each of the standard audiology testing frequencies. This too is a remarkable acheivement, with only 12 or 24 variables needing adjustment instead of the usual 200-300 variables found in most other hearing correction systems.
Either way, we stand apart in the world with our proprietary tuning systems. Nobody else has anything like it for the quality of corrections we can offer.
Very exciting stuff!
Q: Can you explain a bit about your calibration procedure?
A: Our calibration procedure has absolutely nothing to do with your hearing. Rather, it has to do with establishing known sound intensity levels throughout the sound system. We follow the procedures broadly outlined by Bob Katz, and most frequently aim for a standard K-14 RMS-Wide system of calibration.
To do this, we disable the processing of the System 5000 (or any other system of ours), but we leave in place the amount of Attenuation needed for when processing is active. Then we pipe a known reference signal, typically a -14 dBFS 1 kHz sinewave, through the sound system. The meters shown on our GUI's should be pointing straight up under this signal condition. That represents 0 dBK-14. At that point we also adjust the output volume level of the headphone amplifier to read 83 dBSPL through a calibrated sound level meter. For speaker systems, a stereo feed should show 86 dBSPL while a mono feed would show 83 dBSPL. These intensities correspond to Dolby/DTS Theater standards for sound playback systems.
As long as you abide by the K-14 system and recognize that sustained levels above the red zone of the meters (+4 dBK14) are unsafe, you will be completely ear-safe while listening to the sound system. Loud crescendi of symphonic movements will sometimes reach as high as (stereo) 90 dBSPL (+4 dBK14 on each meter) but not for any sustained duration.
Once the system has been properly calibrated for K-14 playback, you can enable the processing of System 5000 and know that it will be performing proper spectral warpings.
Now while these are earsafe levels, many people find them too loud for comfortable listening. That's fine. Once we have the system calibrated in this manner, you can lower the output volume of the System 5000. The volume level must be dropped either at, or ahead of, the System 5000 so that it can recognize your reduced playback levels and can compensate properly for them.
For a K-20 calibration, the procedure is essentially the same, except that we utilize a -20 dBFS sinewave. The sound level meter should be using a slow C-weighted scale. We will often double check our results by using a copy of Bob Katz's calibrated -20 dBFS RMS Pink Noise recording.
Many audio professionals are competent to perform this calibration on their own without our help.
The audiology tuning must come from your own audiology measurements made either on your own, or by a professional audiologist. But be aware of the fact that just because our sliders may show 70 dB boost, you will never receive that much gain from any of our systems.
We implement ear-safety limiting for properly calibrated systems to ensure that you never get exposed to dangerous levels. A 70 dB hearing threshold elevation may only translate into a 12 dB boost during music playback. It all depends on the loudness of the music in that Bark band. At no time will you ever see more than 24 dB of boost.
A 70 dB threshold elevation means that in order to hear a sound at the audiological threshold you would need a boost of 70 dB. Hence, with a limit imposed of 24 dB max, you won't be able to hear threshold level sounds. But music never contains such quiet passages anyway.
Q: Is there really an advantage to using de-emphasis followed by outboard analog re-emphasis?
A: This is really a mixed bag...
Our processing inevitably produces some amount of intermodulation distortion, or IMD. We go to great lengths to minimize this, but as we are frequently behaving as a nonlinear compressor in each of the Bark bands, it is inevitable.
Using an outboard equalizer to re-emphasize our de-emphasized boosts affects not only the desired freqeuency boosts but also the inadvertant IMD products. IMD products are found both above and below the signal frequency of a strong carrier. The outboard equalization is generally a high-shelving boost, so those IMD products above the band will be boosted higher.
If we did not perform de-emphasis and did not use outboard re-emphasis equalizers, then these IMD products would not receive any extra boost. That could be a good thing, especially if these IMD products are near audible thresholds.
But, not performing de-emphasis inside the System 5000 means that in some cases of severe hearing corrections, we will have to use up as much as 24 dB of available processing dynamic range.
That too, is probably okay, since we are operating on 24-bit audio streams. 24 dB represents 4 bits, so we would be left with 20-bit dynamic range in the music. And frankly, this is about the practical limit for any currently available, even high-quality, D/A converters. So using up 4 bits of internal dynamic range will likely be veiled by the output to analog anyway.
In the end it really depends on the preferences of the listener. Some people prefer the outboard re-emphasis equalization, and others don't care either way. The difference, of course, can translate into a huge cost difference.
Clearly, without the availability of 24-bit audio this would be an easy choice. To forego the outboard equalization in a 16-bit system means you would end up with 12-bit audio. That would sound horrible, but 20-bit audio probably sounds about as good as you can ever get.