Laryngectomees are a clinical population who, typically due to cancer, have their larynx surgically removed. As a result the patient has to live without vocal folds. One possible method to regain the ability to speak is the usage of an artificial electronic larynx (EL), which is a mechanical device that produces sound vibrations that are transmitted through the neck tissue into the vocal tract. Such EL-speech, is far from sounding natural, which results from the nonlinear transducer design of today’s available ELs and the very limited prosody.

In this thesis we examine a variety of sound signal generation approaches as potential speech source models, that could be used in a future EL, based on a linear transducer design, that is simulated in this thesis by an electrodynamic exciter. The speech quality resulting from the selected and implemented models is then evaluated by using a Comparison Category Rating (CCR), trying to explicitly assess the perceived frequency spectrum, the listening effort, the noise and the overall quality. The results indicate a improvement for some models in the naturalness of the produced speech which is in direct association with the perceived frequency spectrum.