Pitch Estimation of Speech Signals Based on the Energy Ratio of Multi-Channel Correlation

Pitch estimation is the most basic and important task in speech signal processing, and it has been applied in many speech applications. This paper proposes a novel pitch extraction algorithm for speech signals based on a time-domain signal processing approach, from the point of view of the similarity between multi-channel signals. The proposed algorithm and several classic pitch estimation methods are evaluated on the TIMIT and CSTR databases, considering various acoustic noise conditions with different degrees of non-stationarity and signal-to-noise ratios. The experimental results demonstrate that our proposed algorithm achieves the lowest average error rates and mean error values in most cases.