Search

This report describes the design, principles of operation, and performance characteristics of an Advanced Development Model of a voice recognition system (VRS) which can serve to input cartographic data to a computer. The completed system has been installed at the Defense Mapping Agency Aerospace Center (DMAAC) at St. Louis, MO, for evaluation and testing. The VRS is intended for use in entering by voice cartographic data to the Digital Landmass System (DLMS) Data Base. It was designed to satisfy the DMAAC product specifications. The software developed for the VRS includes two complete stand-alone programs. Performance tests conducted at TTI disclosed an average system word recognition accuracy of just under 99 percent for five talkers. The recognition tests were conducted by the use of tape recordings. These tape recordings were made during a previous contract involving cartographic data entry. Each person spoke approximately 536 words after uttering five training repetitions. The test results were virtually identical to those obtained during the previous contract.

This report describes the configuration and testing of a large vocabulary voice recognition system which could be used for image interpretation and intelligence report generation involving the creation of descriptive statements with unpredictable content and variable format. Recently, Threshold Technology developed an innovation in voice input technology which combines keyboard input with automatic voice recognition to provide accurate and rapid entry of a large vocabulary of words with no requirements for structuring the data entry process. This innovation, referred to as TALK & TYPE, is a system in which the user types the first letter of each word to be entered while he speaks the word into a word recognition device. The typed letter then activates only those vocabulary words which start with that letter, thereby greatly reducing the size of the working vocabulary and greatly increasing recognition accuracy and speed. (Author)

This report describes the design, principles of operation and performance characteristics of an Advanced Development Model of a limited-vocabulary voice recognition system. This system is intended to demonstrate the use of voice data entry in cartographic applications such as the processing of Digital Radar Landmass Simulator (DRLMS) data and the preparation of Flight Information Publications (FLIPS) data. Three separate operating computer programs were supplied for system demonstrations. The first program, the DRLMS Fixed Node program, is designed to simulate the manner in which voice could be used to input DRLMS FAC data to a cartographic computer. The second program simulates a possible mode of inputting data for the construction of FLIPS terminal charts. The program is a variable node program with a total vocabulary size of 500 words. This structure of this program allows recognition of a maximum of 100 words in any node of a maximum of 30 nodes. Node structures may be easily changed. The cartographic word recognition system is based upon the Threshold Technology Inc. (TTI) commercial VIP-100 isolated-word recognition system. Performance tests conducted at TTI disclosed an average system word recognition accuracy of 99% for talkers experienced with word recognition systems.

Speed and accuracy have been compared for isolated-word voice recognition, keyboard, and graphical menu data entry systems. One entry task involved simple copying of numeric and alphanumeric data strings. A second was a simulation of complex flight data entry scenario. The factors evaluated included voice response feedback and prompting, hand occupation during data entry, and subject experience. Keyboard provided the fastest and most accurate entry of numeric data strings and the fastest entry of alphanumeric strings by subjects with keyboard experience in the simple scenario, but was slow relative to voice and graphical menu for entry of words by inexperienced subjects in the complex scenario. Voice entry provided the lowest error rate for entry of alphanumeric data strings in the simple scenario primarily because of its greater immunity to reading errors. In the complex scenario, voice was faster than keyboard for inexperienced subjects, and had a similar operational error rate, but had a substantially higher error rate before correction. Graphical menu ranked between keyboard and voice in most of the simple scenario measures, except that it was least accurate with alphanumeric data, and had the lowest entry speed for long strings.

This report describes a 2-part program (1) to interface an existing word recognition system (WRS) to a bathymetric digitizer and (2) to develop a new word recognition system for a large vocabulary. The first program provides an improved means for entering bathymetric readings from smooth sheets to punched cards. The WRS allows a cartographer simultaneously to obtain X-Y coordinate locations and provide voice data entry of depth reading for each coordinate location. With the operator's hands free to concentrate on the X-Y position sensing cursor he can speak the depth numbers. If they are correctly recognized, he can enter them onto punched cards without losing sight of the smooth sheet. The cartographic WRS which was interfaced to the digitizer is described in AD-B008 303L. The complete system has been installed at the Defense Mapping Agency Hydrographic Center at Suitland, MD. The large vocabulary WRS will be used to explore other cartographic applications. This system is based upon the VIP-100, a WRS system with capability of recognizing up to 200 separate words in a syntactic structure. The structure allows up to 30 words in any node of the structure.

This report describes the development, operation and performance characteristics of an improved Advanced Development Model of a Voice Input Code Identifier (VICI). The VICI is an isolated word recognition system capable of recognizing the English digits and four control words, CANCEL, ERASE, VERIFY and TERMINATE. The improved VICI system is based upon equipment previously developed under Contract F30602-74-C-0171. This original VICI system was intended for use by male talkers only and required wide bandwidth speech data input. The advanced development model is capable of operation with input speech which has been limited to telephone bandwidth. It will accept digits and control words as spoken by either male or female talkers. No training of the system by a speaker is necessary. By the use of an alpha-numeric output display, a speaker using the system can verify that each digit spoken into the system was correctly recognized. Errors can be corrected through the use of the control words. In addition, the system can detect and correct many single digit errors which occur within six digit code groups. This error correction algorithm uses check digits which are included in the code groups to detect errors. The system then corrects the errors when possible or requests a reentry of the data by the talker. A speaker dependent software package also was developed which provides for the recognition of up to 200 words.

This report describes the design, development and feasibility evaluation of a key word recognition system designed to detect key words/phrases embedded in narrow bandwidth conversational speech. It has been a goal that the techniques developed during this program be independent of the speaker. A real-time, dual processing level, key word recognition technique has been developed during this program. The technique utilizes the occurrence of a sequence of phonemic events to identify a candidate key word. A fine grained analysis is then performed on the speech segment containing the candidate word to determine if it is the key word of interest. A series of narrow band data recordings have been generated during the course of this investigation. An improved sequential logic element has been developed and the features and programmable sequential logic system of the RADC Voice Information Processor (VIP) has been interfaced to a minicomputer.