The exemplar work looks like a good approach, as Jort has provided Matlab code, I would imagine that it could be readily tested for your particular case.
Another approach to think about, if you have a section of isolated noise in your audio recording, is the pioneering work by Ephraim & Malah.
They produced the statistical method of spectral noise estimation and removal ... the trick in your case would be to make sure that you are using an analysis window which is long enough to be able to capture the low frequency components of your noise source.
You can find a SPDemo application in the downloads section on their web page :
On 09/20/2012 12:19 AM, Jort Gemmeke wrote:
I dont have experience with MRI scanner sounds in particular, but it seems like such a structured interfering source should be a good match for the exemplar-based source separation techniques I have been using over the past few years. You can find more information on my website, www.amadana.nl , as well as a demo (matlab) and some links to related techniques that may be of use.
Feel free to contact me directly if you think this can be of use, then we can give it a try.
Jort GemmekePostdoctoral ResearcherKU Leuven, ESAT-PSI Speech group
Date: Wed, 19 Sep 2012 16:59:06 -0400
From: Kyle Jasmin <kyle.jasmin.11@xxxxxxxxx>
Subject: Subtracting regularly repeating sounds
Content-Type: text/plain; charset=ISO-8859-1
I wonder if anyone has successfully removed the sound of a helium pump from
auditory recordings made in an MRI scanner. It occurs regularly (every 1s),
and changes over time (Praat thinks it is a formant with downward
trajectory). Any advice is appreciated on subtracting regularly repeating
sounds would be appreciated.
Speech Communication Laboratory, UCL Institute of Cognitive Neuroscience
& Laboratory of Brain and Cognition, National Institute of Mental Health