Updated MINDS Report on Speech Recognition and Understanding, Part 2
Janet M. Baker, Li Deng, Sanjeev P. Khudanpur, Chin‐Hui Lee, James Glass, Nelson H. Morgan, Douglas D. O’Shaughnessy · 2009
psychological aspects of human speech perception include the essential psychoacoustic properties that underlie auditory masking and attention. Such key properties equip human listeners with the remarkable capability to cope with “cocktail party” effects that no current ASR techniques can successfully handle. Intensive studies are needed in order for ASR applications to reach a new level, delivering performance comparable to that of humans. Specific issues to be resolved in the study of how the human brain processes spoken (as well as written) language are the way human listeners adapt to nonnative accents and the time course over which human listeners reacquaint themselves with a language known to them. Humans have amazing capabilities to adapt to nonnative accents. Current ASR systems are extremely poor in this regard, and improvement is expected only after we have sufficient understanding of human speech processing mechanisms. One specific issue related to human speech perception (and linked to human speech production) is the temporal