Dual-pitch separation for harmonically-related pitches and noisy sounds

Abstract

The third generation of pitch detectors based on physiological and psychological knowledge has advanced to a stage that two pitches can be identified. Such pitch detectors are usually based on the estimate-cancel-estimate paradigm to iteratively detect a desired number of pitches in a sound. A problem associated with this paradigm is that it does not know when to stop because the number of salient pitches in a sound may not be known in advance.

This thesis describes a dual-pitch detection model based on Meddis and O'Mard's psychophysically faithful methods for extracting pitch. Its main strengths over Meddis and O'Mard's model include the capability to handle noise and harmonically-related pitch components.

Date of Award	2002
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology