A study of F0 estimation based on RAPT framework using sustained vowel

Acoustic analysis plays a vital role in the study of human voice. Acoustic parameters such as Fundamental frequency (F0) can be derived directly from voices. It is one of the most proven parameters that determines the normality or abnormality of voice disorders in people. The Statistical measures of fundamental frequency, such as mean F0 and standard deviation F0 are suitable for clinical assessment of voices. In this study, these measures are computed using Robust Algorithm for Pitch Tracking (RAPT) and Dr. Speech algorithms. The common error that occurs in the fundamental frequency estimation algorithm is the octave error. The post-processing method namely de-step filter is applied to rectify this error. This study uses Dr. Speech as the benchmark and evaluates the robustness of the RAPT algorithm for the original F0 contour and de-step filtered contour. The results indicate that the use of the filter gives significant improvement to the performance of the RAPT algorithm, thus making it suitable for clinical evaluation of the fundamental frequency.