Voiced/unvoiced determination of speech signal in noisy environment using harmonicity measure based on instantaneous frequency

This paper presents a voiced/unvoiced determination algorithm using instantaneous frequency amplitude spectrum (IFAS) in adverse environments. The proposed algorithm measures the degree of periodicity of speech signal, defined as harmonicity measure, where the difference between voiced part and unvoiced speech can be quantitatively obtained. We describe a new technique for voicing decision using an IFAS-based F/sub 0/ evaluation function with variable window length and IF band selection. The proposed technique is evaluated with a speech signal corrupted by additive white Gaussian, pink, and traffic noise. The results show that the proposed method outperforms ESPS, AMDF and TEMPO for both female and male speakers in all simulated conditions.