There are many factors which affect the performance of automatic speec
h and speaker recognition systems. On its journey from the human to th
e machine the speech signal encounters several sources of corruption.
These include articulation effects such as stress, environmental effec
ts such as noise, and channel effects inherent in the telephony networ
k. The recognition task itself introduces another dimension of difficu
lty. Although isolated-word recognition is relatively straightforward,
natural language input remains a formidable challenge. Five areas of
research have been selected which illustrate the diverse field and dem
onstrate the key issues which must be addressed in order for the techn
ology to flourish.