Browse Prior Art Database

Language Operated Pattern Recognition Decision Implementer

IP.com Disclosure Number: IPCOM000087008D
Original Publication Date: 1976-Nov-01
Included in the Prior Art Database: 2005-Mar-03
Document File: 5 page(s) / 79K

Publishing Venue

IBM

Related People

Dixon, NR: AUTHOR [+2]

Abstract

A major problem in pattern recognition is the segmentation of data complexes in the time domain. The problem obtains in areas such as processing of electrocardiographic, electroencephalographic, seismic, sonar, radar, and speech data. While the present approach is applicable to any of the above, or specific interest is the segmentation of continuous-speech data approximately at the phonemic level.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 42% of the total text.

Page 1 of 5

Language Operated Pattern Recognition Decision Implementer

A major problem in pattern recognition is the segmentation of data complexes in the time domain. The problem obtains in areas such as processing of electrocardiographic, electroencephalographic, seismic, sonar, radar, and speech data. While the present approach is applicable to any of the above, or specific interest is the segmentation of continuous-speech data approximately at the phonemic level.

There are two major ways of approaching this problem. The first involves modeling of the processes involved and the derivation of results based on output of the model. The second involves the use of heuristic rules. The present approach represents an attempt to afford convenient generation, implementation and modification of essentially heuristic rules. A general language-operated decision implementation system (GLODIS) is an operating system which consists of a user-oriented language (GLODIS), a language compiler and an implementer. This implementer, while constrained for structural simplicity and ease of data management, maintains its functional generality.

Given a set of N simultaneous time sequences, or "lines of data", a general pattern recognition problem, using these data as input, is to determine an output time sequence which specifies each time point as belonging to some class. In the usual case, this means the segmentation of time into regions of varying size, with, perhaps, some label affixed to each region. This is the goal of the system described herein, which has been designed to accommodate phonemic segmentation of English speech signals.

The object of the decision procedure is to determine an output sequence which indicates the desired segmentation. In complex situations this sequence is not obtainable in a direct, one-step manner; rather, a sequence of decisions is necessary based upon some hierarchy of confidence. Speech segmentation is an example where this complexity is needed. The GLODIS approach stresses flexibility in the generation, implementation and modification of complex decision procedures; it is an operating system.

Fig. 1 shows the basic structure of this operating system. The user control of the system, written in GLODISL, is fed to a compiler stage, the function of which is to convert the user-oriented specifications in GLODISL to a set of ordered control data for the implementer. Input data, in the form of time-aligned sequences, are also fed to the implementer. The heart of GLODIS is the implementer, which applies the control data to the input data in making the decisions necessary for determination of outputs.

Figs. 2 and 3 show the basic flow of the GLODIS implementer. An implementer with great flexibility can be constructed in any number of ways. It is necessary to strike an acceptable balance between simplicity and generality of operation. In accordance with this need, a set of solidifying, but not especially limiting constraints are introdu...