Hi Qin,
The brainstem neural circuits that are thought to analyze the high-frequency pinna cues in the HRTF for sound elevation and front-back disambiguation are in the dorsal cochlear nucleus
(DCN). For a paper on how the DCN is likely processing the rising and falling edges of spectral notches, see
https://doi.org/10.1523/JNEUROSCI.4963-04.2005.
There are a number of groups that have developed models of the DCN circuitry, but these are fairly old now. See the results of this search on Google Scholar for a list of such models:
https://scholar.google.com/scholar?q=computational+model+dorsal+cochlear+nucleus
Note that these studies have primarily been focused on replicating the physiological data for DCN cells, rather than for estimating sound source location from HRTF filtering. In
contrast to the MSO circuit, it is unlikely that there is a simple mapping from the DCN model response to an estimate of the sound source location – it is likely that machine learning is required to learn the spectral cues for a given individual’s HRTFs.
If you are just interested in front-back disambiguation rather than sound elevation estimation, and you are okay with using signal processing methods rather than a physiological
model, then this fairly recent paper provides a good overview of some of the algorithms that are available:
https://doi.org/10.1186/s13636-021-00235-2.
I hope this helps.
Ian
Ian C. Bruce,
PhD, PEng
Professor
Department of Electrical and Computer Engineering
McMaster University
1280 Main Street West
Hamilton, Ontario, Canada L8S 4K1
location:
ITB-A213
phone:
(905) 525-9140 x 26984
email:
brucei@xxxxxxxxxxx
web:
http://www.ece.mcmaster.ca/~ibruce/
Google Scholar
ORCID
From: AUDITORY - Research in Auditory Perception <AUDITORY@xxxxxxxxxxxxxxx>
On Behalf Of Qin Liu
Sent: March 3, 2025 5:44 AM
To: AUDITORY@xxxxxxxxxxxxxxx
Subject: Re: [AUDITORY] Seeking advice on using ANF firing rate to reslove front-back confusion in sound localization model
|
Some people who received this message don't often get email from
000003c563e12bd3-dmarc-request@xxxxxxxxxxxxxxx.
Learn why this is important
|
|
Caution: External email.
|
Dear Sathish,
Thank you for your warm response, and I apologize for my delayed reply.
Your understanding is correct. Currently, I can only predict whether the sound source is in the front or back hemifield.
Could you please suggest any models or papers that I can refer to, which detail how to incorporate cross-frequency cues to accurately determine or classify whether a sound source is in front or behind?
Thank you very much for your help.
Best regards,
Qin
From: Sathish Kumar <sathish.sreeni58@xxxxxxxxx>
Sent: Wednesday, 26 February 2025 13:35:31
To: Qin Liu
Cc: AUDITORY@xxxxxxxxxxxxxxx
Subject: Re: Seeking advice on using ANF firing rate to reslove front-back confusion in sound localization model
Dear Qin,
I agree with Prof. Jan Schnupp that some pre-assumptions are necessary for your problem. In one of my research projects, I simulated the auditory nerve responses of spatialized signals using the zillany2018 model, and their Direction of
Arrival was estimated using the dietz2011 model. However, I found that the responses were interpretable only for fine structure data, not envelopes.
From my understanding, the Dietz model considers only sound sources within the frontal hemifield. For front-back localization, you could use cross-frequency integration to detect the high-frequency attenuation caused by the pinna, allowing
you to determine whether the sound source is in front or back hemifield. This can then be followed by the Dietz model to estimate the DoA, as ITD and ILDs are nearly identical for front and back azimuths.
Best regards,
--
Sathish Kumar
PhD Research Scholar
Department of Audiology and SLP
Kasturba Medical College (MAHE), Mangalore
Mob: 9789447666 || E-mail:
sathish.sreeni58@xxxxxxxxx