ETCvérif, a prototype of a cooperative automatic speech recognition system
Jean-Luc Cochard, Philippe Langlas, Gérard Chollet · The Journal of the Acoustical Society of America · 1994
ETCvérif a speech processing environment under development at IDIAP, serves as a testbed for experimental design issues in the development of robust and efficient speech recognition applications. This works stems from the strong intuition that a probable solution to the general problem of speech understanding lies in the development of a system able to deal cooperatively with a large set of distinct, partial and even unreliable problem solvers, namely HMMs, NNs, APD (acoustic phonetic decoder), prosodic analyzers, and parsers. This paper will present the general architecture of ETCvérif focusing upon the main properties of ETC (environment for cooperative treatment). In particular, the core of the system is devoted to execution scheduling of speech and language processing modules; communication and negotiation over partial, compatible but not unifiable results. The design decisions regarding the requirements of cooperation between (modules such as) phonetic HMMs, an APD with lexical access and a bottom-up parser will be described as part of a prototype of speech recognition presently under realization. [Work supported by the Swiss National Funds for Scientific Research.]