It is widely accepted that motor and auditory processes interact in speech perception, but little is known about the functional role motor processes play in the development of speech perception. To address this question we consider a Bayesian model of speech perception development based on three sets of variables: motor representations M, sensory representations S and objects O (e.g. phonological units such as phonemes). The model comprises two internal branches. Firstly, an auditory identification sub-system connects S and O. Secondly, a motor sub-system connecting M and O and a sensori-motor model connecting M and S can be combined to provide “motor identification” of sounds S, from S to M and from M to O, in an analysis-by-synthesis process. Development is modeled as a learning process in which a master iteratively produces a sensory percept S associated with an object O. The learning agent updates its auditory sub-system by observing S and O. Update of the two other branches is more complex and based on an imitation phase. The learning agent estimates a likely motor action M from input S, produces this M and observes the resulting sound S’. M, S’ and O are used to update both the motor sub-system (M, O) and the sensori-motor model (S, M). We show that the auditory identification sub-system learns rapidly, and becomes efficient for stimuli close to those provided by the master, although it generalizes poorly. By contrast, the two other sub-systems evolve more slowly, and in consequence the motor identification system performs less accurately. However, motor identification happens to have captured more variable situations during learning, and generalizes better (e.g. in noise). This is in line with a developmental schedule in which auditory processing is mature before motor knowledge (Kuhl et al, 2008) and is exploited by infants after 11 months of age for analysis-by-synthesis of unusual speech stimuli (Kuhl et al., 2014).