The aim of this work is to provide a set of robust references which can be used for the phonetic annotation of both normal and pathological speech. The approach adopted is to make use of sensors of speech productive activity which give outputs which are of auditory, contrastive, value. An apparatus for multi-sensor recording and the physiological characteristics of the sensors in responding to speech production are described. Signal characteristics for different speech sound categories and speakers are presented. The performance of the sensors on continuous speech is assessed. Finally, a framework for the use of the sensor signals for (semi-)automatic annotation and labelling of dialogue is described.