ESCA Tutorial and Research Workshop on
This paper provides an overview of the Recogniser Sensitivity Analysis (RSA) methodology, under development by Logica within the UK Alvey Speech Technology Assessment (STA) project MMI/132, for assessing the performance of automatic speech recognition systems, both as products and as algorithms under development. The applicability of the method for predicting likely field performance in a particular application as well as its diagnostic capability will be discussed. Preliminary results indicate a very strong correlation between certain parameter values and recogniser performance as well as the potential for predicting field performance with limited data. The paper concludes with a discussion on future directions for extending RSA.
Bibliographic reference. Peckham, Jeremy / Thomas, Trevor / Frangoulis, E. (1989): "Recogniser sensitivity analysis: trial results and future directions", In SIOA-1989, Vol.2, 119-126.