Renato De Mori

Parallel Algorithms for Syllable Recognition in Continuous Speech

A distributed rule-based system for automatic speech recognition is described. Acoustic property extraction and feature hypothesization are performed by the application of sequences of operators. These sequences, called plans, are executed by cooperative expert programs. Experimental results on the automatic segmentation and recognition of phrases, made of connected letters and digits, are described and discussed.

Computer recognition of plosive sounds using contextual information

The design of the plosive consonant recognition component of a speech understanding system is described. The recognition rules used incorporate information on coarticulatory and multiple cue effects. The rule combination and scoring system is a fuzzy model of the human speech perception and information integration mechanism. These rules can be also used in classical stochastic models or dynamic programming approaches to automatic speech recognition.

An expert system for mapping acoustic cues into phonetic features

Abstract The paper describes the conception of the sublexical levels of a speech-understanding system as a society of experts. Experts cooperate in extracting and describing acoustic cues, generating and verifying phonetic hypotheses, and accessing a large lexicon. The knowledge of each expert is described by a frame language which allows integration between structural and procedural knowledge. Structural knowledge deals with relations between facts like acoustic-cue descriptions and phonetic-feature hypotheses. Procedural knowledge deals with rules for the use of relations, for the generation of contextual constraints for relation application, and for the extraction of new cues in specified contexts. The main purpose of the research proposed here is that of providing at the same time a model for computer perception and algorithms useful for designing complex systems operating in real time. Some experimental results on the performance of the proposed system are reported.

Extraction of acoustic cues using a grammar of frames

The conception of Speech Understanding Systems (SUS) as perceptual models is useful for designing machines capable of performing more ambitious tasks than the ones actually achieved by commercial machines and laboratory prototypes, offering the challenge of experimenting new parallel algorithms and non-conventional system architectures. Central to the organization of an SUS is the representation of knowledge structured on several levels of abstraction and the control strategy that has to use the knowledge efficiently. This paper announces a new knowledge-based system for interpreting speech patterns jn a taskindependent multi-speaker environment. Central to the conception of this system is the idea that the interpretation of speech patterns is controlled by rules applicable to segments of the speech signal having approximately the duration of a syllable. This is consistent with the conclusions drawn from experiments on the storage time of a pre-perceptual auditory memory. In contrast with other approaches which use syllables as units for speech recognition or for lexical access, the generation of syllabic hypotheses is not based on matching spectral segments of speech with prototypes. Rather, it is controlled by rules that take into account context-constraints, bot tom-up information and top-down predictions imposed by lexical hypotheses. Rules also control the extraction of acoustic cues which may depend on the hypotheses to be generated or verified. The use of lexical-dependent constraints in syllable hypothesization is consistent with perception models. Syllable hypothesization is based on the hierarchical application of relations involving acoustic cues extracted from the signal or spectral transformations of it and phonetic features. The selection of relations to be applied is controlled by a planning system described by a frame language. This language is introduced for representing knowledge about how to interpret the speech signal and its transformations. It describes the available knowledge and how it can be used and updated. Knowledge is shared, in the model proposed in this paper, among Experts. They cooperate in extracting acoustic cues from the signal and its numerical transformations, in generating hypotheses about bounds of syllabic segments and about phonetic feature hypotheses inside the segments. A high degree of parallelism can be achieved with such a model allowing to design real-time systems with complex tasks. The knowledge of each expert can be updated, particularly for what concerns the specification and the use of contextual constraints whose importance has been stressed in previous works. The attempt to conceive an SUS as a knowledge-based system is also on the line of recent researches in Computer Vision in the hope of finding good models for Computer Perception. The system proposed here is based on rules which capture many speaker-independent relations between phonetic features and acoustic cues. These rules are a kernel that can be enriched when a new knowledge is acquired regarding new pronounciations and new languages.

A recursive algorithm for binary multiplication and its implementation

A new recursive algorithm for deriving the layout of parallel multipliers is presented. Based on this algorithm, a network for performing multiplications of twos complement numbers is proposed. The network can be implemented in a synchronous or an asynchronous way. If the factors to be multiplied have <italic>N</italic> bits, the area complexity of the network is <italic>O</italic>(<italic>N<supscrpt>2</supscrpt></italic>) for practical values of <italic>N</italic> as in the case of cellular multipliers. Due to the design approach based on a recursive algorithm, a time complexity <italic>O</italic>(log <italic>N</italic>) is achieved.nIt is shown how the structure can he pipelined with period complexity <italic>O</italic>(1) and used for single and double precision multiplication.

Design for a recursive parallel multiplier

A network for performing multiplications of two twos complement numbers is proposed. The network can be implemented in a synchronous or an asynchronous way. If the factors to be multiplied have N bits, the area complexity of the network is O(N2) for practical values of N as in the case of cellular multipliers. Due to the design approach based on a recursive algorithm, a time complexity O(log N) is achieved.

On the Structure and Use of Task-Dependent Knowledge

Task-dependent knowledge contains syntax, semantics, and pragmatics. Its purpose is the solution of ambiguities and recovery from errors which still remain at the lexical level. Furthermore, task-dependent knowledge may be used to transform the surface structure of a recognized sentence into its conceptual representation.

Parallel Algorithms for Hypotheses Generation in Continuous Speech

The paper describes the conception of the auditory, syllabic and lexical components of a Speech Understanding System (SUS) as a Society of Experts.

On the Use of Syntactic Pattern Recognition and Fuzzy Set Theory

The scheme introduced in Section 4 of Chapter 1 represents an active model for understanding speech.

Rules for Characterizing Sonorant Sounds

The purpose of this chapter is to introduce and describe the rules which may be applied when the feature “sonorant” is hypothesized for a segment.