Related Experiment Videos
A spectral network model of pitch perception
M A Cohen1, S Grossberg, L L Wyse
1Center for Adaptive Systems, Boston University, Massachusetts 02215, USA.
The Journal of the Acoustical Society of America
|August 1, 1995
Summary
The spatial pitch network (SPINET) model simulates pitch perception by transforming spectral sound data into pitch strengths. This new model explains complex pitch phenomena like octave shifts and dominance regions effectively.
Area of Science:
- Auditory Neuroscience
- Computational Acoustics
- Psychoacoustics
Background:
- Previous spectral models of pitch perception have limitations in explaining a broad range of perceptual pitch data.
- Understanding the neural mechanisms of pitch perception is crucial for auditory processing research.
Purpose of the Study:
- To develop and analyze a novel neural network model for pitch perception, the spatial pitch network (SPINET) model.
- To integrate spectral pitch modeling with basic neural network signal processing for enhanced simulation capabilities.
- To interpret model components as peripheral auditory processing stages for potential application in sound source separation.
Main Methods:
- The SPINET model transforms spectral representations of acoustic sources into spatial distributions of pitch strengths.
- A weighted "harmonic sieve" is employed, where pitch activation strength depends on weighted sums around harmonics, with lower harmonics having greater influence.
- An on-center off-surround network is utilized for noise suppression, partial masking, and edge pitch phenomena.
Main Results:
- The model successfully simulates pitch perception data, including mistuned components, shifted harmonics, and continuous spectra like rippled noise.
- Harmonic weighting functions explain the dominance region and octave shifts in pitch perception for ambiguous stimuli.
- The model accounts for pitch perception in Shepard tone complexes and Deutsch tritones without requiring attentional mechanisms.
Conclusions:
- The SPINET model provides a comprehensive framework for understanding pitch perception, integrating spectral and neural processing.
- The model's components offer insights into peripheral auditory processing and can be integrated into larger architectures for complex auditory tasks.
- SPINET demonstrates the ability to predict various pitch phenomena, highlighting the importance of harmonic weighting and neural network architectures.