For those with neurological or developmental disorders compromising speech, brain machine interfaces could help them communicate. But today's interfaces are slow and, from electrodes placed on the scalp, can detect letters only. The speech generated is robotic and affectless. Neuroscientists have now shown that they can reconstruct the song a person is hearing from brain recordings alone, holding out the possibility of reconstructing not only words but the musicality of speech, which also conveys meaning.