A cascade autocorrelation model of pitch perception

Area of Science:

• Auditory perception
• Computational acoustics
• Psychoacoustics

Background:

• Autocorrelation algorithms and auditory periphery models are established for pitch prediction.
• Certain complex stimuli challenge the predictive accuracy of existing autocorrelation models.

Purpose of the Study:

• To investigate if modifications to peripheral auditory models and autocorrelation algorithms can maintain pitch prediction accuracy.
• To adapt existing computational pitch models to accommodate challenging auditory stimuli.

Main Methods:

• Extended an existing auditory model with a low-pass filter for within-channel autocorrelations.
• Incorporated nonlinear basilar membrane processing and adaptive integration time constants.
• Evaluated the modified model using diverse, problematic stimuli like click trains and transposed tones.

Main Results:

• The enhanced model demonstrates improved pitch prediction for stimuli that previously posed difficulties.
• Successfully predicted pitches for short tone sequences, mixed-interval click trains, and shuffled click trains.
• Showed reasonable success with transposed tones, a known challenge for autocorrelation methods.

Conclusions:

• Modifications to auditory periphery models and autocorrelation algorithms can preserve and enhance pitch prediction capabilities.
• The extended model offers a more robust approach to computational pitch perception for complex auditory signals.
• This research addresses limitations in current models, paving the way for more accurate auditory processing simulations.