The hand-held electrolarynx (EL) voice prosthesis is widely used for restoring voice and speech after the removal of the human larynx due to trauma or disease. The goal of this thesis is to implement electromyographic (EMG) control to improve the EL. The EMGcontrolled EL (EMG-EL) provides hands-free control over voice initiation, termination, and fundamental frequency modulation. The EMG-EL was evaluated through real time experiments and computer simulations. EMG-EL performance was then optimized by surgical modifications to provide a vocal-related signal source, systematic training to improve voluntary EMG control, and advanced control strategies to produce less discrepancies between EMG-EL voice and normal voice timing.

In order to obtain voice-related EMG signals, eight subjects underwent a nerve transfer procedure during total laryngectomy. The recurrent laryngeal nerve (RLN) was transferred to denervated neck strap muscles on one side in all eight subjects (RLNstraps), and the strap muscles were preserved with their normal ansa cervicalis innervation on the other side in five of the subjects (ansa-straps). Neck surface EMG was measured periodically for over a year after the surgery. Signals correlated with speech production were obtained in all subjects from both the RLN-straps and the ansa-straps. After one year of post-surgical recovery, the RLN-strap EMG signals exhibited larger magnitudes and more precise timing than ansa-strap signals during intended phonation.

In a voice reaction time experiment, seven normal subjects produced EMG-EL voice initiation that was as fast as normal and manually-controlled EL voice, and voice termination that was as fast as the manually-controlled EL voice. After a 10-day training protocol, four other normal subjects produced fluent speech with effective intonational stress variations. Furthermore, training of three laryngectomy subjects with RLN-straps led to fluent EMG-EL speech production in all subjects, and successful EMG-EL pitch modulation for intonational contrasts in two of the three subjects.

Computer simulations of the EMG-EL on/off triggering using recorded EMG signals showed that the use of an advanced dual threshold control strategy significantly improved the predicted performance in all cases. Furthermore, EMG-EL simulations confirmed that the RLN-strap signals provided performance that was superior to both the ansa-strap signals of the laryngectomy subjects, and the intact strap muscles of ten normal subjects.

In conclusion, muscle EMG signals can be successfully used to control an EL device, thus enabling total laryngectomy subjects to produce hands-free alaryngeal speech with dynamic pitch modulation. The performance of the EMG-EL users can be optimized through the use of RLN-innervated neck strap muscles, and further enhancements can be expected through systematic training and more advanced EMG processing and EL control strategies.