To predict the performance of a probabilistic speech recognizer it is often desirable to estimate the contribution of the language model and that of the acoustic model. We describe an approach to this problem which tries to take into account the interaction between the two sources of information. Some results are presented concerning the 20000-word vocabulary, real-time IBM recognizer of the Italian language.
Bibliographic reference. Ferretti, Marco / Maltese, Giulio / Scarci, Stefano (1989): "Measures of language model and acoustic model information in probabilistic speech recognition", In EUROSPEECH-1989, 2473-2476.