For decades speech recognition has been assumed to rely on frequency cues, but recent research has demonstrated the saliency of temporal cues. By studying how different forms of speech are encoded in the peripheral auditory system, we have identified temporal cues that are invariant across speaker and means of production. These cues work for the recognition of normal, whispered or even noise-vocoder speech. The identification of these cues is important for developing better cochlear implants, hearing aids and automatic speech recognition systems.