A highly parallel single-instruction multiple-data (SIMD) array signal processor is advocated as efficient for a wide range of real-time problems. We examine its performance for digital speech recognition and show that impressive throughput rates for realistic vocabulary sizes can be achieved for ‘time-warping’ dynamic programming algorithms which currently form the basis of several commercial and research speech recognisers.
References
-
-
1)
-
J.S. Bridle ,
M.D. Brown ,
R.M. Chamberlan
.
Continuous connected word recognition using whole word templates.
Radio & Electron. Eng.
,
167 -
175
-
2)
-
Roberts, J.B.G., Darby, B.J.: `Signal processing architectures for the VLSI age', Military Microwaves Conf. Proc., 1982, p. 277–286.
-
3)
-
Moore, R.K., Russell, M.J., Tomlinson, M.J.: `Locally constrained dynamic programming in automatic speech processing', Proc. IEEE Int. Conf. ASSP, 1982, p. 1270–1272.
-
4)
-
D.J. Hunt ,
S.F. Reddaway ,
G.S. Scarrott
.
(1983)
Distributed processing power in memory, The fifth generation computer project.
http://iet.metastore.ingenta.com/content/journals/10.1049/el_19830690
Related content
content/journals/10.1049/el_19830690
pub_keyword,iet_inspecKeyword,pub_concept
6
6