This paper deals with the sound quality of electro-larynx devices, which is one method of communication for people who have lost their larynx. Current commercially available devices are characterized by an unnatural, mechanical sound. Assuming the availability of a linear transducer several alternative excitation signals are compared to the sound of a state-of-the-art electro-larynx. The signals considered are both physical models and waveform models. In a listening test 10 sentences, recorded by two healthy electro-larynx speakers using the different excitation signals were evaluated by 20 listeners. Results suggest that a more natural speech sound may be possible without sacrificing intelligibility.