The present study aimed to determine whether observation of upper limb actions selectively influences speech production. We compared the effects on children with those on adults, hypothesizing that action observation is used by children for speech learning. Children and adults observed an actor either grasping a cherry or an apple, or bringing the same fruits to his mouth. They pronounced the syllable/ba/ at the end of the action. In a control experiment, children and adults executed the two bringing-to-the-mouth actions, still pronouncing/ba/. As previously found ([Euro. J. Neurosci., 17 (2003) 179]; [Euro. J. Neurosci., 19 (2004) 192]), the observed kinematics of the action, which were different according to the fruit size, influenced lip shaping kinematics and voice formants. In addition, the effect was selective for the action since the observations of actions such as grasping and bringing-to-the-mouth affected formant 1 and formant 2 in the voice spectra, respectively. The effects on speech were greater in the children than in the adults. By contrast, the effects on lip shaping did not differ between the two groups. Effects similar to those found for action observation were found for action execution in spite of a different arm kinematics between children and adults. The results of the present study are discussed according to the hypothesis that action observation induces in the viewer action recognition and activation of the successive mouth act (probably grasping-with-the-mouth when observing grasping-with-the-hand and chewing when observing bringing-to-the-mouth). This subsequently seems to affect characteristics peculiar to the emitted vowel. This mechanism might have been used by humans to transfer a primitive arm gesture communication system from the arm to the mouth and may be further used by children for speech learning.