Evaluation of a method for vowel-specific voice source control of an electrolarynx using visual information

The electrolarynx (EL) is a widely used device for alaryngeal communication, but the low quality seriously reduces the intelligibility of EL speech. To improve EL speech quality, a vowel-specific voice source based on visual information of lip shape and movements and artificial neural network (ANN) is implemented into an experimental EL (SGVS-EL) system in real time. Five volunteers (one laryngectomee and four normal speakers) participated in the experimental evaluation of the method and SGVS-EL system. Using ANN participants were able to perform high vowel precision with identification rates of >90% after the training. The results of voicing control indicated that all subjects using SGVS-EL could achieve good vowel control performance in real time, but still control errors frequently occurred at the voice initiation period. However, the control errors had no significantly impact on the perception of SGVS-EL speech. Intelligibility evaluation demonstrated that both the vowels and words produced using the SGVS-EL were more intelligible than vowels spoken with a commercial EL (by 30%) or words (by 18%), respectively. Using a controlled vowel-specific voice source was a feasible and effective way to improve EL speech quality with more intelligible words.