Comparison of Fo extraction methods for high-precision voice perturbation measurements

Voice perturbation measures, such as jitter and shimmer, depend on accurate extraction of fundamental frequency (Fo) and amplitude of various waveform types. The extraction method directly affects the accuracy of the measures, particularly if several waveform types (with or without formant structure) are under consideration and if noise and modulation are present in the signal. For frequency perturbation, high precision is defined here as the ability to extract Fo to +/- 0.01% under conditions of noise and modulation. Three Fo-extraction methods and their software implementations are discussed and compared. The methods are cycle-to-cycle waveform matching, zero-crossing and peak-picking. Interpolation between samples is added to make the extractions more accurate and reliable. The sensitivity of the methods to different parameters such as sampling frequency, mean Fo, signal-to-noise ratio, frequency modulation, and amplitude modulation are explored.