Researchers trained a neural network to map audio “voiceprints” from one language to another.