This paper provides a review of recent developments in speech recognition research. The concept of sources of knowledge is introduced and the use of knowledge to generate and verify hypotheses is discussed. The difficulties that arise in the construction of different types of speech recognition systems are discussed and the structure and performance of several such systems is presented. Aspects of component subsystems at the acoustic, phonetic, syntactic, and semantic levels are presented. System organizations that are required for effective interaction and use of various component subsystems in the presence of error and ambiguity are discussed.

Speech recognition by machine: A review

Hearsay II (HSII) is a system currently under development at Carnegie-Mellon University to study the connected speech understanding problem. It is similar to Hearsay I (HSI) in that it is based on the hypothesize-and-test paradigm, using cooperating independent knowledge sources communicating with each other through a global data structure (blackboard). It differs in the sense that many of the limitations and shortcomings of HSI are resolved in HSII. The main new features of the Hearsay II system structure are: 1) the representation of knowledge as self-activating, asynchronous, parallel processes, 2) the representation of the partial analysis in a generalized three-dimensional network (the dimensions being level of representation (e.g., acoustic, phonetic, phonemic, lexical, syntactic), time, and alternatives) with contextual and structural support connections explicitly specified, 3) a convenient modular structure for incorporating new knowledge into the system at any level, and 4) a system structure suitable for execution on a parallel processing system. The main task domain under study is the retrieval of daily wire-service news stories upon voice request by the user. The main parametric representations used for this study are 1/3-octave filter-bank and linear-predictive coding (LPC)-derived vocal tract parameters [10], [11]. The acoustic segmentation and labeling procedures are parameter-independent [7]. The acoustic, phonetic, and phonological components [23] are feature-based rewriting rules which transform the segmental units into higher level phonetic units. The vocabulary size for the task is approximately 1200 words. This vocabulary information is used to generate word-level hypotheses from phonetic and surface-phonemic levels based on prosodic (stress) information. The syntax for the task permits simple English-like sentences and is used to generate hypotheses based on the probability of occurrence of that grammatical construct [19]. The semantic model is based on the news items of the day, analysis of the conversation, and the presence of certain content words in the partial analysis. This knowledge is to be represented as a production system. The system is expected to be operational on a 16-processor minicomputer system [3] being built at Carnegie-Mellon University. This paper deals primarily with the issues of the system organization of the HSII system.

Organization of the Hearsay II speech understanding system

An organization is presented for implementing solutions to knowledge-based AI problems. The hypothesize-and-test paradigm is used as the basis for cooperation among many diverse and independent knowledge sources (KS's). The KS's are assumed individually to be errorful and incomplete.

A uniform and integrated multi-level structure, the blackboard, holds the current state of the system. Knowledge sources cooperate by creating, accessing, and modifying elements in the blackboard. The activation of a KS is data-driven, based on the occurrence of patterns in the blackboard which match templates specified by the knowledge source.

Each level in the blackboard specifies a different representation of the problem space; the sequence of levels forms a loose hierarchy in which the elements at each level can approximately be described as abstractions of elements at the next lower level. This decomposition can be thought of as an a prion framework of a plan for solving the problem; each level is a generic stage in the plan.

The elements at each level in the blackboard are hypotheses about some aspect of that level. The internal structure of an hypothesis consists of a fixed set of attributes; this set is the same for hypotheses at all levels of representation in the blackboard. These attributes are selected to serve as mechanisms for implementing the data-directed hypothesize-and-test paradigm and for efficient goal-directed scheduling of KS's. Knowledge sources may create networks of structural relationships among hypotheses. These relationships, which are explicit in the blackboard, serve to represent inferences and deductions made by the KS's about the hypotheses; they also allow competing and overlapping partial solutions to be handled in an integrated manner.

The Hearsay II speech-understanding system is an implementation of this organization; it is used here as an example for descriptive purposes.

/pdf/a-multi-level-organization-for-problem-solving-using-many-15mxgvwkrw.pdf

A multi-level organization for problem solving using many, diverse, cooperating sources of knowledge

The authors detail an integrated system which combines natural language processing with speech understanding in the context of a problem solving dialogue. The MINDS system uses a variety of pragmatic knowledge sources to dynamically generate expectations of what a user is likely to say.

High level knowledge sources in usable speech recognition systems

Speech is man’s most natural channel of communication, so it is only natural that it should be the subject of much work in pattern recognition. Recognizing speech is difficult, for the nature of the speech communication process is such that the resulting acoustic signal is highly encoded and full of contextual dependencies. Speech perception on the part of a human listener requires him to utilize his linguistic knowledge to undo the encoding. Machines that aspire to recognize the same signal must be able to apply similar constraints. Current work in automatic speech recognition ranges from systems that recognize isolated words from limited vocabularies to systems that “understand” the meaning of naturally spoken sentences, using higher level linguistic constraints in addition to the information derived from the speech signal itself. This chapter describes the techniques developed and the progress made in speech recognition and understanding in the early 1970’s.

Speech Recognition and Understanding

This paper describes an implementation strategy for a vocal data management system (VDMS) being developed by the voice input/output project at the System Development Corporation. VDMS will accept connected speech of a language describable by 25-50 phrase equations and having a vocabulary of approximately 1000 words formed from about 100 data records. The strategy is based on the concept of predictive linguistic constraints (PLC). The present concepts of fixed directionality in parsing are replaced by a more generalized approach. To facilitate this flexibility, the system comprises a set of near-independent coroutines that are interconnected by a software busing structure. The VDMS acoustic processors verify the predictions. Very loose matching criteria are used for locating the predicted words. Special attention is given to word segments that are experimentally determined to be most invariant.

/pdf/a-vocal-data-management-system-djxya819cp.pdf

J. Barnett

Papers

A vocal data management system