When speakers produce sounds they need to have an "image" of a sound before they can produce it: They need to know what the sound they are going to produce shall be like. This "image" of the sound could consist of articulatory and/or acoustic components. For a sound such as /f/, for example, speakers could have an articulatory representation such as "constriction between lower lip and upper incisors". However, they could also have an acoustic representation such as "noise with a diffuse spectrum". The study investigates whether in perturbed speech speakers compensate with the aim to produce the same articulatory configuration as in unperturbed speech or the same acoustic output. If speakers adapt towards the same articulatory configuration this would support articulatory images of sounds. If speakers head towards a certain acoustic output one could assume that they have acoustic images of sounds. The results of this EMA-study show that during early compensatory attempts speakers use articulatory images, later, however, they change the articulation while focussing on the acoustic output. Final compensation efforts are directed towards an optimisation of the movement.