Re: Timing Comparison?

I spent a few years searching around for the boundaries of my “Crescendo” Algorithm. When I began, I thought I would perform corrections every 1/4 Bark, with each band spanning 1/2 Bark.

That works, but it was a beast to run on any computer at the time, ca. 2005. I was also attempting to solve as faithfully as possible the complete EarSpring-Conductor equation system to derive corrections. All this on a system aimed at live performance use. The computers of that time simply could not do it. I had to resort to DSP’s.

Since that time, I found that to be massive overkill. And in order to make the algorithm run on a cell-phone, we had to drop back to 5 bands (mono), and use approximations on the loudness solutions. That was for street use, mostly human conversations.

And since those early days, computers today outstrip the DSP’s of yesterday. I can now do stereo audio processing at 192 kHz, all in high-level Lispworks Lisp.

For good musical listening, my current system in the lab here is using 11 bands, each 2-Bark wide. And it sounds amazingly good to me. The band centers are identical to the standard audiology testing frequencies. 

My thinking at the time was that, since those frequencies are the only ones measured in an audiology exam, clients of the Crescendo system might want to tune their own systems more critically than the offered 1-knob solution that I described earlier. 

But that really hasn’t turned out to be the case. Most of use, myself included, are quite happy with the 1-knob tuning, with identical corrections for both left and right ears. Our brains interpolate from one side to the other.

My work in developing Crescendo also developed an algebra for hearing corrections, and as a result, I am uniquely capable of correcting conditions of Hyper-recruitment (aka HyperAcusis) and Decruitment, in addition to normal Recruitment hearing. It turns out that many (most?) of us have all 3 conditions in addition to some bands of normal hearing.

Audiology testing is only good to ±5 dB, and that assumes they are calibrated correctly. (I highly doubt it in most cases.) But the technical writeup shows that errors in measured threshold levels as great as 5 dB will surely produce inferior results. And it turns out that if you have severe enough impairment, your tests are almost guaranteed to overestimate your hearing loss. And you can quite will find your own to better than 1 dB using the Crescendo system.  If you don’t have severe impairment, then the ±5 dB will be just fine.

Our hearing simply doesn’t have loudness discernment much better than that. And so by using these fewer bands, we get a smoothed, or interpolated, response across adjacent bands.

I am thrilled every time I put on my headphones and listen to music through Crescendo. Because I inhabit a darker world, I notice when unusual sounds occur - sounds that most people learn to discard - such as the buzzing of the harp strings on the Harpist’s fingernails, and the knocking of her knuckles against the wooden frame. It is almost like I now have Golden Ear hearing with Crescendo.

- DM

> On Nov 3, 2025, at 13:58, Toomas Altosaar (as toomas dot altosaar at gmail dot com) <[email protected]> wrote:
> 
> Am enjoying this discussion on psychoacoustics. I coded my first Bark spectral analyzer (48 channels, 0-24 Bark, 1/2 Bark step resolution, 5 ms time step, fs was 22.05 kHz) on our Symbolics 3670 during the summer of 1987.
> Wish our acoustics lab would have met up with Dave’s work, it blows me away!
> 
>> On 3. Nov 2025, at 16.11, David McClain (as dbm at refined-audiometrics dot com) <[email protected]> wrote:
>> 
>> sheesh… typing corrections??
>> 
>> Within a critical band, you cannot separately distinguish the loudness of two nearby tones. Or rather, the loudness masking from the louder sound drowns out the presence of the lesser sound.
>> 
>> At any rate, a true-to-life correction system would need this kind of spectral fluidity. But for use in a realtime correction system, intended to help performers in live sessions, you can’t dilly-dally. You have to respond very quickly, on the order of 5 ms. And so you have to live with approximations.
>> 
>> Our loudness response is not nearly as critical as our pitch perception, and so approximations work well enough.
>> 
>>> On Nov 3, 2025, at 07:05, David McClain (as dbm at refined-audiometrics dot com) <[email protected]> wrote:
>>> 
>>> I would say a LPF is too crude. It is all sound level dependent. At low sound levels you do have a severe lowpass filter effect - my own are woofers with -24 dB/octave above 1 kHz.
>>> 
>>> But at loud enough levels the curve flattens our to normalcy.
>>> 
>>> So this makes the system a nonlinear compressor, in each Bark band.
>>> 
>>> Bark bands are really not fixed in frequency - they are self organizing around the loudest spectral peaks, and their bandwidth describes the critical bands of hearing. Within each critical band, you can separately discern the loudness difference between two nearby tones.
>>> 
>>> The nonlinear compression curves look a lot like the J of a ski-jump. They level out at low sound levels, and then rise steeply to become nearly normal at loud levels. Level-out means they must apply amplification in inverse proportion to the dB level of incoming sound. Steeping means they need to apply ever less amplification. At the very loudest levels, our hearing all converges to normal.
>>> 
>>> When you have Sensioneural impairment - as contrasted against a U-shaped hearing curve often seen among former Artillery Officers - then you have impairment at all frequencies and the degree of impairment is nearly a straight line in dB measure against Bark frequency. The variation from one person to the next is just the degree of impairment, but their line is parallel to those of other people in dB-Bark space.
>>> 
>>> 
>>>>> On Nov 3, 2025, at 06:56, Yuri Davidovsky <[email protected]> wrote:
>>>> 
>>>> 
>>>> 
>>>>> On 3 Nov 2025, at 14:40, David McClain <[email protected]> wrote:
>>>>> 
>>>>> What we found is that all humans have impairment slopes that average around 3.2 dB/Bark. What varies is the degree of impairment, but not its spectral slope.
>>>> 
>>>> So in essence the main finding is that hearing impairment is essentially modelled by a low pass filter curve? Meaning, it predominantly affects the higher frequencies, rather than manifesting itself by a total signal attenuation?
>>>> 
>>>>> 
>>>>> So, once you adjust to the degree of hearing impairment at, say 4 kHz, the all other levels of impairment correction can be deduced from that degree and the known spectral slope.
>>>> 
>>>> Do you mean, the hearing impairment curve tends to start at a specific band, so what you need is to find the band at which the attenuation starts and compensate the attenuation curve that affects the rest of the spectrum? Since as you said, that impairment slopes tend to be similar among individuals.
>>> 
>>> 
>>> _______________________________________________
>>> Lisp Hug - the mailing list for LispWorks users
>>> [email protected]
>>> http://www.lispworks.com/support/lisp-hug.html
>> 
>> 
>> _______________________________________________
>> Lisp Hug - the mailing list for LispWorks users
>> [email protected]
>> http://www.lispworks.com/support/lisp-hug.html
> 
> _______________________________________________
> Lisp Hug - the mailing list for LispWorks users
> [email protected]
> http://www.lispworks.com/support/lisp-hug.html


_______________________________________________
Lisp Hug - the mailing list for LispWorks users
[email protected]
http://www.lispworks.com/support/lisp-hug.html