Learning new words involves consolidation. After one night's sleep, not only is explicit knowledge about the novel words enhanced, but the new words also now compete with similar-sounding existing words (Dumay & Gaskell, 2007) during word recognition. The present study assessed whether lexical consolidation strips off surface details of newly learned words, producing more abstract representations. We manipulated the speaker's voice between exposure and test. Participants learnt one set of novel competitors (such as 'shadowks' for 'shadow') seven days before the test, and another set immediately before the test. Each word was learnt in a male or a female voice, and was tested in either the same or the other voice. Cued recall and phoneme monitoring showed stronger memory for the seven-day old items and, if anything, an enhanced voice effect (i.e., better performance in the same voice condition) after seven days. Crucially, our most indirect measure of lexical competition showed that only the seven-day old items (as expected) engaged in lexical competition, but only when the input preserved the voice in which they had been encoded. These findings indicate that consolidation does not make word representations more abstract: voice specific details do not just survive lexical consolidation; they are enhanced by it.