Separation and estimation methods of harmonic and noise signals in voice sources using independent component analysis (ICA)

Based on the physiology and pathology of vocal phonation, the oral speech, glottal volume velocity waveforms obtained by two-pass method for glottal inverse filtering analysis, and voice signals transmitted from the skin above thyroid were separated into harmonic and noise signals using independent component analysis (ICA). And furthermore, the harmonic-to-noise ratios were estimated and compared with the results obtained by conventional methods in the cases on different vowels, phonation types and several typical laryngeal diseases. The optimum method estimating the harmonic-to-noise ratios in normal and pathological voices were validated, respectively, which is of great importance to the vocal physiology, phonetics and diagnoses for laryngeal diseases.