I spent the night reimplementing my experimental version of Crescendo, to allow for detailed audiometry at the standard test frequencies, and further incorporated the hyper-recruitment & decruitment adjustments.
To be most correct, the pre/post-Crescendo gains need to be applied in Phons space, not in the usual SPL space with which we are all familiar. There are subtle, and not so subtle, differences in this kind of Phons equalization.
I was encouraged by the experiments on my decruitment hearing discussed in the previous post. I wanted to know how different (better?) it might sound if a proper equalization in Phons were applied. Indeed it sounds great! A bit more subtle that the boosts provided by a simple conventional high-shelving EQ, but tantalizingly better on the very highest frequencies.
The control panel shows the best estimate by a professional audiologist at the bottom of the pane. Left and Right channels are treated separately. And beneath each row of the audiology is a row for gain adjustments – positive values for decruitment, negative for hyper-recruitment. At the moment, I’m not using any hyper-recruitment corrections.
The Crescendo engine takes those audiology measurements, which are in dBHL (SPL space) above a presumed threshold level in SPL. It then converts those internally into Phons threshold elevations, using the ISO-226 model.
The decruitment adjustment acts as a post-gain on Crescendo, in Phons space, and also decreases the Phons threshold elevation by the same amount. This avoids excess noise floor magnification while also operating on a gentler slope in the recruitment curve around threshold levels.
If hyper-recruitment were indicated, with negative gains, then it applies the gain ahead of Crescendo rendering, and leaves the Phons thresholds alone.
But in all cases, we are operating in Phons space, not SPL space. And the difference is delicious.
I don’t know why I ignored my decruitment for so long, but I’m very glad that I finally paid deep attention to it.
VTuning (a made-up term) attempts to model hearing impairment as a progressive exponential degradation, highest at the highest frequences, and declining by about 3.3 dB/Bark toward lower frequencies. The VTuning value is supposed to be the dBHL threshold elevation for 4 kHz.
This slope was found by comparing a number of audiograms mapped into Bark/Phons space and seeing that there was general agreement with a straight line slope among the collection. Each audiogram exhibits a different depth of impairment at 4 kHz, but if you align them they all tend to line up along that line.
However, this is a simplified model of sensioneural hearing loss. There are other effects that can be seen, and my own audiogram displays them. For example, I have a pronounced 10-15 dB plateau excess above that straight line at frequencies from 1-4 kHz.
So the current experiment looks at the damage around 1.5 kHz, and attempts to equalize all the growth in thresholds as due to decruitment above that initial level. For example in my left ear I show a 55 dBHL impairment. Rather than abruptly correcting at 2 kHz with a 4 Phon decruitment, I ease into it by giving half that amount to the 1.5 kHz level, for an implied elevation of 53 dBHL.
Then at each higher frequency I add whatever amount of decruitment will make the audiology match that 53 dBHL. The right ear is treated similarly, but with its own lesser impairment level at 1.5 kHz.
This isn’t quite correct either, because the audiometry is in dBHL (SPL space) while the decruitment gains are in Phons. But this is the current experiment. The errors are hopefully small. I will look more closely at the expected errors in a moment…
But I must say, after listening through both VTuning and the decruitment-corrected audiometry, that decruitment treatment definitely sounds more correct to my ears. Speech sibilance is readily audible to me with the decruitment treatment, but just barely present with only VTuning. Sibilance is mostly around 6-7 kHz.
So, doing the decruitment in this manner, keeping the base audiometry impairment level at all higher frequencies, is roughly in keeping with the following reasoning:
As the sound storm enters the cochlea, the greatest damage will occur at the very highest frequencies. And as that storm dissipates energy as it moves further along the basilar membrane, we expect less damage going toward lower frequencies. A freshman-level physics problem is to state the differential equation for an effect which dissipates in proportion to its remaining size. The solution, of course, is exponential decay – or in dB measure, a straight line.
But since my level of damage is severe, it may well have happened that the hair cells became calcified and less willing to move. That might account for the decruitment. It takes more sound power to get them moving, but the actual underlying sensory damage corresponds to a milder threshold elevation in the absence of the stiffening.
Hence, calcification should likewise be worse at the highest frequencies, for hair cells nearest the oval window. I imagine the underlying sensory damage to be the result of inoperative hair cells “screaming” at the top of their lungs, and hence masking sounds beneath the elevated threshold, sensed by the remaining undamaged hair cells. There is only a tiny fraction of hair cells in the inoperative condition.
This discussion of physiology is mere musing on my part. Not a serious explanation of the damage. But the 1-4 kHz plateau threshold excess is very real, and this is one possible explanation. Listening to the music with decruitment corrections tells me that I am on the right track here.
Interestingly, I’m finding no need for dual-Crescendo engines now. A single Crescendo engine running on the indicated audiology with decruitment corrections fits the bill quite nicely.
A more refined approach attempts to split the increase in threshold at higher frequencies between the kind of sensory damage found with vTuning, and that caused by stiffening of the hair cells. The vTuning model shows a growth rate of about 3.3 dB/Bark, and our bands are roughly 2.5 Bark wide, so we should see about 8 dB increase in sensory threshold as we step higher in frequency. That almost corresponds to the 50/50 partitioning with decruitment gains shown here:
It is hard to tell if this is better. Perhaps… the extreme highs are a bit less attention grabbing, but it still sounds very good.