Voiced and Unvoiced Sound Detection Based on Wavelet Transform

Abstract

Voiced-unvoiced detection is a fundamental research topic in speech signal processing, holding significant importance for applications such as speech recognition. Addressing the differences in frequency-domain energy distribution between voiced and unvoiced sounds, this paper proposes a voiced-unvoiced detection method based on discrete wavelet transform. By performing multi-level wavelet decomposition on speech signals, the method extracts energy features from high-frequency wavelet coefficients at different scales. Combined with an energy threshold decision strategy, it achieves the distinction between voiceless and voiced consonants. Experimental results demonstrate the method's effective differentiation between voiceless and voiced consonants, showing feasibility under the selected experimental conditions.