[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: NAL-1 or similar hearing aid simulator?


About half a year ago I completed (as co-researcher and principal
implementor) the new NAL-NL2 prescription algorithm for the National
Acoustics Laboratories.

I am not aware of anything which is available for free out there,
however I can describe a method for implementing it yourself. The
NAL-NL2 prescription algorithm will be available to research groups in a
month or few, and it will be reasonably priced for research groups and
small companies... to wet your appetite, it based on an enhanced
loudness and speech intelligibility model and a far more advanced method
of prescription synthesis from arbitrary hearing loss profile.

So once you have acquired the new NAL-NL2 (or other) prescription
algorithm, what do you need to do ? I will specify it for you liberally
in the hope that you are willing to implement it.

As it is based on psychoacoustics, you need two paths... one path is the
adaptive acoustic path, the other path is the control path which guides
the adaptation...

a] The adaptive acoustic path can be implemented in the time domain as a
set of IIR filters or in the Fourier (frequency) domain as an FIR filter
- either way, you are targeting a maximum latency of roughly 3ms.

b] The control path is what they loosely call a 'compressor' in industry
- it assesses the dB SPL of the signal over different sub-bands and
using the prescription, derives the correct gain to apply to each IIR
filter OR DFT bin.

I will give more detail here ...

a] Assume you are using the Fourier (FIR) approach ... because your
filter is adapting, you will want to use waveform (or windowed) overlap
add. For each 3ms window of time, you will convolve (multiply) your DFT
with the correct gain prescription. The foundation of psychoacoustic
hearing aids are really that simple ... others like to add 'features'
which improve the hearing aid and try to make it more realistic ... this
is where we are right now ... industry is spending millions on small %
improvements ... the foundation (which I have just mentioned) generally
gives you - on average - your largest speech intelligibility

b] The foundation of the control arm is also rather simple, it is
commonly called a 'compressor' and it consists of the following cascaded
i] A level estimation - simple power in dB SPL for each sub-band of
ii] A time dependant average of the signal power in each sub-band - this
is as simple as a low pass filter with two time constants ... one
shorter time constant for rising power levels and one longer for falling
power levels.
iii] A gain matrix. The gain matrix is your prescription. It is rated at
each incremental power level for each sub-band.

Now what happens ? Well, think of it in terms of one sub-band at a
time... (i) The power level in that sub-band is (ii) smoothed and the
(iii) prescription is indexed... You now know what to convolve your
acoustic window with in order to improve the speech intelligibility for
the hearing impaired.
Again, various other things may be added at this point for
'fancification', however as before, the largest percentage of speech
intelligibility improvement is achieved using this technique. Every
other addition will (on average) only improve your hearing aid by a
small number of percent. By the way, any new (tiny) improvement may only
give you on average a few % improvement in speech intelligibility and in
general will cost you millions to get in a product.

I have some very strong opinions on the current state of the art of
hearing aids ... and most of them are based on the following
observations ...
a] Hearing aids started as linear amplifiers. (Acoustics)
b] Hearing aids became nonlinear amplifiers. (Psychoacoustics)
c] NOTHING has changed in that respect - since the beginning.
d] This is why the hearing aid industry is spending millions on
obtaining an extra few percent of speech intelligibility improvement,
over what we currently call the state of the art.

We need to step into a new paradigm of processing for the hearing
impaired. The first company to do this with a completely new processing
strategy and paradigm may well outperform the current state of the
art... this is the danger to current groups/companies who don't step
back and look at the big picture with a clear head ...

Whilst I think current hearing aids (which are ALL based on
psychoacoustics) are rather clunky and cumbersome, no-one has put
forward an alternative solution.

I have something in the pipe line who's signal processing is purely
physiologically based ... I am looking for interested parties to help me
get it off the ground ... it is currently my number one goal ...


On Sat, 2010-06-12 at 10:32 +0100, Richard - UK wrote:
> Hi,
> Has anyone come across a PC utility which can take an audio file and a
> hearing loss description file as inputs, in order to produce a
> 'corrected' output audio file?
> Thanks!
> Richard