This study explores how much speech can be temporally compressed and still understood by blind people who have daily practice with speech synthesis vs. sighted persons without such training. Texts with formant-synthesized speech, and compressed natural speech with and without pauses were generated at rates between 9 and 14 syll/sec (sighted persons) and 17 and 22 syll/sec (blind). The removal of pauses in compressed natural speech shows significant benefits at only few speaking rates. Results also show that synthesis is understood worst by sighted but best by blind. The fact that some of the blind still understood speech at 22 s/s reveals the flexibility of speech perception during the processing of ultra-fast speech.