Roon DSP Procedural EQ/Convolution To Aid Sound Induced Hearing Loss / Hearing Impairment

Roon Software Discussion
21

In response to @Maxxim_M comments.

So where would you start, @Maxxim_M. Given your earlier enthusiasm, I don’t think the answer is “I wouldn’t” but perhaps it is.

I’ve been operating on the assumption that if we could take the 9 data points per ear and both a) interpolate within the tested range and b) extrapolate to expand the range, that we could create a reasonable approximation of a curve to use as a starting point.

Once we have the curve, determining offset directionality (peak vs. dip) and magnitude is clearly not as simple as adding or subtracting decibels to get to a target. I get that. Intuitively, and given your clarification of the non linearity of hearing models, it seems that a non-linear adjustment based on deviation from the mean may be starting point. Maybe.

So the first problem is interpolation and I was going to play with spline interpolation.

The second problem is extrapolation and you’ve suggested that simply extending the edges out might have side effects or, worse, do damage.

Any thoughts on these? Suggestions for approaches?

The third problem becomes determining offsets and I guess the simple question there is “If we’ve been successful at the solving the interpolation and extrapolation issues, and we’ve produced a reasonable approximation of a curve, can that curve be used to generate effective vectors and, if so, how?”

You asked for questions…there they are :slight_smile:

If you’re already tired of this, I will absolutely understand. I appreciate your input to this point.

3 Likes
22

Whatever you do, don’t shoot your ears any further. Since I’m not an expert, you don’t have to listen to me :slight_smile:

23

Hello.
@gTunes, we actually have at least two solutions to the problem;

  1. The scientific route
  2. The experimental route

If we focus on the goal of improving the perception of the musical experience, it doesn’t matter which path you take, the important thing is the effect of whether you are satisfied at the end.

Personally, I’m in favour of the scientific route (maybe it’s because that’s my design and I like to know what I’m doing) Your approach written in three points is good and correct. You can extend it with accurate audio test data. The more data the more accurate the fit and the better the end result. ( I will use the example of shoes that are too loose or too tight, both versions are limiting but walking is possible :wink:
Personally, I would start by setting a target.
I would carry out all possible tests of peripheral hearing, neural hearing and auditory processing ( it doesn’t hurt to have them done myself ).
Convert the collected results into units that I can use in sound gating, equalization.
I would compare the results by superimposing the initial curves with the results after corrections.
I would check the effect organoleptically. ( If I don’t hear what I should hear I have made a mistake somewhere).
Basically a simple graph of the instruments sound production is sufficient for this. At what frequency what is produced/reproduced via I hear, I don’t hear
This is how I see the beginning of the journey.
( it’s never so good that it couldn’t be better )

The experimental path:
Basically you use whatever is available and at hand to your satisfaction. Twist, shift, add, subtract. There are as many possibilities as there are knobs on your equipment! If you are happy with the result, it’s OK! It’s only a subjective evaluation, but it’s important that YOU are satisfied.
It is not necessary to have specialist knowledge in order to participate in a discussion on a given topic. I am inspired by questions, because something that seems obvious to me, may not be so for my colleagues. Therefore, I will disagree with @Marian who wrote ; “Since I am not an expert, you don’t have to listen to me …”. I have to yes I do have to listen because for me it is the knowledge of how I can help to explain and maybe be inspired by a simple idea to solve a problem. That is why I thank you @Marian.
By the way in answer to your last question. No I don’t have enough questions I am happy to read I will listen I will discuss :wink:

Greetings
Maxxim

1 Like
24

Hello,
I created a longer posting about this issue here:

I used a microphone (Umik1) and REW to measure what I do not hear. Like a 13khz wave that I see in REW but do not hear that. I then crank up the amplifier until I just hear it sitting at my favorite place and all channeled through my system and speakers. I note down the amplification needed in dB and do the whole bandwidth, not bothering above 13khz.
I then create a REW filter and export that for Roon. Be aware that hearing loss requires steep boosts like 20 db which most filters will not do. Dirac does a maximum of 8 db, as the distortion is too much above. I am not sure what Roon allows, it might be worth checking.

2 Likes
25

The experimental route

1 Like
26

@Marian Excellent advice. Boosting high frequencies to compensate for sensorineural hearing loss could be a slippery slope. My hearing is down considerably staring at 3KHz, but I still enjoy listening. I have had tinnitus for a decade, and now am starting to experience ‘recruitment’, an increased sensitivity to certain frequencies. I worry about further damage to my hearing, and recently introduced PEQ to notch out frequencies that were perturbing. I’ve long given up on the ideal of a perfectly flat response curve equalized to my hearing response. Now it’s about setting the response to only provide an enjoyable experience, and to DO NO FURTHER HARM.

3 Likes
27

This is a perfect summation of my own evolving experience.

Thank you to @Maxxim_M, @Marian, @Joachim_Strobel for all of the fantastic input. The thread @Joachim_Strobel referenced was a fascinating read and a cautionary tale.

I did have some fun going down the data science path. I pulled my audiometry results into a Jupyter notebook. As a starting point, I extended the low and high results out to 250Hz and 2kHz - just added a number of invented data points as @Marian suggested. The I pulled the frequency list of the 695 measured frequencies from AutoEQ, threw those into a text file, and did both linear and and spline interpolations.

For fun, and in case anyone wants to play around with something similar, here are the resulting plots and the code (this is just for one ear).

There’s a pretty straightforward path from this work to an input into AutoEQ if someone wants to pursue that.

Personally, I’ve come full circle on this and am simply using the audiometry data to inform a couple of simple PEQ filters I’ve created. For the moment, I’m simply using different left/right peak curves with a wide Q targeted at the big dip in the midrange (centered at 3.5kHz). With a -2 dB headroom value, I can bump that range up by 6dB on the left and 8 dB on the right without clipping. I’m currently doing this in conjunction with headphone-specific convolution filters (the PEQ after convolution).

Prior to testing, I was using a speaker filter to reduce volume on the left side to equalize what I heard across ears. This approach is better for me. It’s imperfect, but it’s an improvement.

I’m sure I’ll keep chipping away and experimenting/learning. Thanks again for all of the great input.

image
image

import numpy as np
from scipy.interpolate import interp1d
import matplotlib.pyplot as plt

# Original data - x is frequencies, y is measured dbhl....points below 250 and above 8000 are invented
x = np.array([20, 100, 250, 500, 1000, 1500, 2000, 3000, 4000, 6000, 8000, 10000, 12000, 14000, 16000, 18000, 20000])
y = np.array([20, 20, 20, 10, 10, 35, 30, 50, 50, 20, 10, 10, 10, 10, 10, 10, 10])

# Read new x values (frequencies) from frequencies.txt
with open('frequencies.txt', 'r') as f:
    new_x = np.array([float(line.strip()) for line in f])

# Create the interpolation function
interp_func = interp1d(x, y, kind='cubic')
#interp_func = interp1d(x, y, kind='linear')

# Compute new y values using interpolation function
new_y = interp_func(new_x)

# Plot the original data and the interpolated data
plt.plot(x, y, 'o', label='Original data')
plt.plot(new_x, new_y, '-', label='Interpolation')
plt.xlabel('Frequency (Hz)')
plt.ylabel('Hearing Level (dB)')
plt.gca().invert_yaxis()
plt.legend()
plt.show()
1 Like
28

Thanks for mentioning my post in audio science review. For me, the evolving question is the loudness perception. The normal audio tests , doctor or home, tell you when you start hearing a tone. But it does not tell you how you perceive that tone at an overall level of 70db. It is certainly not linear. And the Fletcher-Henderson and ISOxxx curves were taken from young people’s ears. Here is a paper reviewing that topic and shows loudness corrections for elder people. I do believe that the industry should offer such age specific loudness corrections, but who wants to by a receiver for the hearing impaired elder guy:

https://www.researchgate.net/publication/270711273_Comparison_of_equal-loudness-level_contours_between_otologically_normal_young_and_older_adults

3 Likes
29

Great article, the loss of midrange seems to be what I am experiencing and explains why I struggle with overly bright music or systems now as I just don’t have the ability to balance them out. DSP room correction helps but i then do feel like it’s congested at lower volumes and have to crank it up and then it aggravates my tinnitus. I guess it’s also why I can’t focus on conversations when out in a bar or restaurant that’s very noisey and has a lot of reverb.

2 Likes
30

Fascinating article.

It seems that the basic building blocks are there for industry-driven or crowd-sourced solutions. We don’t seem to far away from something like this.

Yes. Testing clarified some of this for me including what’s going on between my left and right ears. The difference is pronounced when using headphones which, as you know, is how I listen most.

During my audiology testing, the audiologist asked me a bunch of questions that I didn’t have context for. Things like “Did your wife send you here?” and “You checked ‘I sometimes have trouble understanding what people are saying’ on the intake form. Was it hard for you to acknowledge that?”

Turns out that she was strategizing for how to introduce the topic of hearing aids - she was gauging my psychological and emotional readiness. I have fairly average sound-induced loss for a man in his mid 50s. But I do have trouble in the contexts that @Simon_Arnold3 describes such as bars/restaurants and movies. So I’m thinking about it.

1 Like
31

Impressive. You don’t mention hearing aids. Do you use them to overcome your hearing loss?

I also have higher frequency drop off and my hearing aids are an essential part of my audio setup. That way, others can share the listening experience without the volume being too high or the EQ being too extreme for them.

2 Likes
32

I have near total loss on one side; bone conduction hearing aids are the only option, but the thought of having something screwed into my skull is discomfiting, never mind the aesthetics.

I also wonder whether it would sound artificial, and how it would affect my perception of audio.

1 Like
33

Coclear implants are a frequent procedure today if indicated by your loss.

1 Like
34

yes; my nerve is fine. problem is with the canal, drum, and middle ear.

this is the type of thing that’s recommended:

35

Many people go to an ENT for their hearing loss. However, a good board certified audiologist has access to a range of aids, and hearing aids have plummeted in price.

1 Like
36

I’m sorry to hear that and I can understand your reluctance. It took quite a while for my hearing to adapt to the hearing aids. It’s a neuro-psychological process aided somewhat by the app on my phone for making fine adjustments to tone and filtering noise.

1 Like
37

thank you. it’s been this way more or less since I was a year old. classic story of ear infections, perforated ear drum (a couple times), surgeries, mastoidectomy (twice)…12 operations in all, although the last one was 30 years ago. options are limited at this point.

1 Like
38

That’s quite a medical history that possibly blighted your childhood and early adulthood. Your determination to find technical solutions to enhance your listening enjoyment is admirable and it appears that you have been broadly successful.

2 Likes
39

Thanks. I haven’t crossed the bridge into hearing aids yet, but no doubt as the ears deteriorate I will have to consider it in the years ahead. Like @woodford I have wondered about the quality of the sound transmission from the hi-fi system through the hearing aid. Something still to research! (I’d be curious if anyone has accumulated experience about this question to share.) Meanwhile, what you say makes sense, in terms of the listening experience of other people around you. For normal ears, the EQ settings that work for me presumably sound much too bright.

3 Likes
40

We’re getting well “off topic” here but nonetheless it is an interesting area where many might wish to contribute or benefit from others’ experience. Maybe someone would like to start a new topic & post a link here?

4 Likes