[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Basilar-Tectorial Membrane - Harmonic Analysis?
As stated in an earlier posting ...
- "What if Source separation produced Source data of a form
that present analysis could not handle? What if Source separation
transforms Binaural data sets into Source data sets and ... those
Source data sets were inconsistent with Fourier analysis?
- Are we ready for a paradigm shift?"
On this line of questioning, I would like to open a discussion concerning
a different view of the Basilar-Tectorial (BT) membrane.
BT resonant response occurs at a given time and actually represents
the largest energy component AT THAT TIME ...
... Flannagan says it well in "Speech Analysis, Synthesis and
Perception", 2nd, Expanded Edition, bottom p149 thru top 150:
"For most speech signals, ... the [BT analysis] provides better
temporal resolution than spectral resolution. Generally, the
only harmonic component resolved mechanically is the
fundamental frequency [highest energy] of voiced segments."
Thus, it is more accurate to think of the Place Theory as producing a
(time) sequence of resonant events. However, a true Fourier analysis
produces a SET of Frequency estimates (and a SET of associated
Phases) at a resonant resolution related to the sampled-time-interval.
A "filterbank" is a (parallel) SET of filters. The BT structure resonants
at only ONE place at only ONE time, i.e., the "filterbank" analogy is
seriously lacking the ability to produce a SET of resonances.
If we agree (?) that the BT is not a "filterbank" then, of what Fourier value
is this single, temporally resonant, BT filter?
The Place Theory predicts particular resonances are transduced by
particular groups of IHC's. But, I've never "heard" discrete, sequential,
sine wave speech, i.e., the ONLY form of speech that could be
*spectrally* detected by a temporal, single-resonance BT detector!
Old 300 baud modems employed this kind of communication.
But, don't be too concerned because BT resonances are NOT the
important feature provided by the BT IHC's ...
... HALFwave transduced time cues are the basic BT data.
But, much more, a whole new form of analysis based on Chaos
(multidimensional geometry) and Self-Organized Neural Maps solves
the Cocktail Party Effect and provides the data streams produced by
separated, acoustic sources.
McGill is running a new version of LISTSERV (1.8d on Windows NT).
Information is available on the WEB at http://www.mcgill.ca/cc/listserv