For a silent speech interface (SSI) to be truly practical, it has to be able to tolerate motion artefacts generated by the user while engaging in normal activities of everyday living. This paper presents a wearable speech restoration system based on magnetic sensors with an integrated background cancellation technique to eliminate the effect of motion-induced interference. The background cancellation technique is assessed when the user makes no intentional movement, when they performs a set of defined head movements and when they make more natural, conversational head movements. The performance is measured for the TIDigits corpus in terms of whole word recognition rate using a Hidden Markov Model and through Mel Cepstral Distortion for a Direct Synthesis of speech using Deep Neural Networks. The results indicate the robustness of the sensing system with background cancellation against the undesirable motion-induced artefacts.