A recognizer of meaningful sequences of isolated words spoken in the Italian language is presented. The system has been designed in a modular fashion and the current version has been obtained by adding the syntactic and semantic levels to the isolated-word recognizer previously developed by the same authors. The lexical, syntactic and semantic levels are integrated into a hierarchical knowledge source represented by a transition network grammar. The parsing is directed by the control unit and is performed by means of a combined strategy of bottom-up and top-down techniques, using a modified version of the Earley algorithm. Lexical hypotheses are verified at the phonemic level assigning them a score based on a distance measure evaluated by an error-correcting parsing algorithm. The protocol used in the preliminary experi - ments is a robot command language.

The paper reports a detailed analysis of the performances ofan Italian-word recognition system structured into a hierarchy of processes controlled by rules of appropriate grammars. The knowledge sources of each recognition level can be learned by experiments and adapted to new vocabularies and new speakers. A set of experiments has been carried out in order to evaluate the influence of vocabulary size and speaker variation over the performances of each recognition step. The aim is to identify a set of parameters whose values can be extracted from a small set of training words spoken by a new speaker and which can be used to automatically adapt the system to the speaker.

The characterization of the influence of lexicon and syntax on the pitch contour in the Italian language is presented. The results of the work will be used in a Speech Understanding System for defining some constraints in the emission of hypotheses and for evaluating the likelihood that a synctatic structure can generate a given pattern of fundamental frequency. The pitch evolution is only determined in voiced tracts; then it is subdivided in a sequence of primitives and translated in symbols. Such strings of symbols are considered as sentences of a set regular languages, whose productions are learned using a stochastic inference procedure. Some results obtained from controlled experiments in which various factors are varied separately are presented.

This paper describes a special-purpose processor for use in performing various operations on sampled signals. The system is fast, flexible, and programmable for performing, in real time, operations such as fast Fourier transformation (FFT) and digital filtering of acoustically derived signals.