BaNa: A hybrid approach for noise resilient pitch detection

Abstract

Pitch is one of the essential features in many speech related applications. Although numerous pitch detection algorithms have been developed, as shown in this paper, the detection ratio in noisy environments still needs improvement. In this paper, we present a hybrid noise resilient pitch detection algorithm named BaNa that combines the approaches of harmonic ratios and Cepstrum analysis. A Viterbi algorithm with a cost function is used to identify the pitch value among several pitch candidates. We use an online speech database along with a noise database to evaluate the accuracy of the BaNa algorithm and several state-of-the-art pitch detection algorithms. Results show that for all types of noises and SNR values investigated, BaNa achieves the best pitch detection accuracy. Moreover, the BaNa algorithm is shown to achieve around 80% pitch detection ratio at 0dB signal-to-noise ratio (SNR).