A speech therapy support tool based on spectral density adjustment

Abstract: The purpose of this work is to expand the toolbox of speech language pathologists and give them the ability to redesign a patient’s voice to a healthier sounding one, thereby giving the patient a target voice to strive towards. A modified voice is generated by applying a finite impulse response (FIR) filter, altering the spectrum of the patient’s voice. In order to facilitate the filter design process, a periodogram is presented with the fundamental frequency and the first few harmonics indicated, estimated with the SWIPE’ method. This way, the therapist gets a visual representation of the voice, which may help with diagnosis. With the filter applied, the patient also gets to hear how a ”healthier” version of their own voice may sound. A survey was conducted containing both modified and unmodified voices. Both therapists and others were invited to participate and evaluate the effectiveness of the modifications. The results show that while naturalness in the voice is mostly unaffected by the filtering, larger modifications are needed in order to perceive difference with regards to criteria such as strain and breathiness.