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

Re: envelope extraction

Hi Ben,
One example of envelope based auditory processing is the cochlear implant. Envelopes are usually extracted from bandpass filter outputs by full-wave rectification and low pass filtering. If it's any use to you, the filter cutoff frequency commonly used for this is 200 Hz. (And some experimental CI processing strategies have used half-wave rectification without any smoothing at all). I'm not sure if there is a neural modeling or physiological reason for the cutoff frequency. 200 Hz retains the temporal modulations of the speech envelope, which some CI listeners are able to hear as a cue to fundamental frequency.
As you say, in normal hearing, neurons can follow much higher frequency stimuli. Although it's not exactly my area, I would assume you would only use a low (e.g. 50 Hz) cutoff frequency if you wanted to model auditory processing without temporal cues. For a physiological model, I don't know of any reason for 50 Hz. In addition to neural refractory effects, there are lowpass cutoff frequencies related to the synaptic delay and re-sequestering of neuro-transmitter by the inner hair cells. But that too is much faster, I think.
Sorry I don't have any references. Maybe someone else can comment.
Daniel Taft

From: AUDITORY - Research in Auditory Perception [mailto:AUDITORY@xxxxxxxxxxxxxxx] On Behalf Of Hornsby, Benjamin Wade Young
Sent: Tuesday, 24 February 2009 3:55 AM
To: AUDITORY@xxxxxxxxxxxxxxx
Subject: [AUDITORY] envelope extraction

Hi all,

I hope the answer to this question is not too obvious but,… We have recently been talking about auditory processing models, in specific, temporal processing models, many of which incorporate an envelope extraction stage. To do this I’ve seen in many cases the use of half wave rectification followed by a low pass filter. The cutoff for this filter is generally a pretty low frequency, say around 50 Hz or so. A colleague suggested that the actual cutoff frequency should be based on the assumed time constant of the system or subsystem being evaluated. I’ve been trying to determine the physiologic rational for such a low frequency filter (slow time constant) in models of auditory processing. Neural processing limitations like the refractory period of the neuron are much faster than this. Anyone have a reference or two that might help explain the physiologic rational for this low frequency filter cutoff?


Thanks much,