Parameters' Extraction and Estimation Methods for Visualization of Telephonic Speech

Abstract

As an aid with which a hearing impaired person can understand telephonic speech, a speech visualization system is being developed. This paper describes methods to estimate speech parameters that include the third formant frequency (F₃) and pitch frequency (F₀). These frequencies are necessary to realize the system, but may not exist in the telephone frequency band.
The third formant frequency is estimated from the first and second formant frequencies by a quadratic equation based on the analysis of multiple regression. For the pitch extraction, we have proposed a method to restore a fundamental component from the harmonic ones using a nonlinear processing and also to measure the periods.
According to the examples of short sentences processed by the above methods, the third formant trajectories and the pitch patterns in telephonic speech almost agree with those in real speech which is recorded at the same time.