This paper presents a brief overview of the speech recognition technology. Algorithms such as Isolated Word Recognition, Dynamic Time Warping, Hidden Markov Modeling, Vector Quantization, Connected Word Recognition, and Continuous Speech Recognition are described. Research in this area is advancing mainly due to resurgence of commercial interest in this technology. Although there are a number of applications that could directly benefit by this technology, a key determinant of success or failure is the human interface. Much research still remains to be done to make this technology robust and usable in real-life applications. Some of these issues and the need for future research are outlined.
This paper describes several algorithms and architectures of digital speech coders. Bit rate compression algorithms based on speech prediction are analyzed for applications such as mobile radio communications, secure voice transmission, voice mail, and voice storage and response systems. The architectures for sample-by-sample or block-by-block processing of speech are discussed. The details of architectures are discussed in a conceptual framework that simplifies the comparison of several implementations of vocoders and the identification of common blocks. Examples of implementation of predictive speech coding algorithms on the TMS320 digital signal processor are presented in this paper. Various predictor structures including new, non-classical, reduced complexity adaptive predictor are presented. Current research trends with an emphasis on low-bit-rate, high-quality speech coding are described.
In this paper results are presented concerning the use of spectral subtraction for enhancing speech corrupted by additive noise. In this study several variations of the spectral subtraction process have been implemented and tested. The overall results provide insight into the use of spectral subtraction for speech enhancement. As a result of this study, several conclusions are presented regarding spectral subtraction, its effects on noise corrupted speech, and its ability to enhance speech intelligibility and listenability. Finally, the use of spectral subtraction as a preprocessor for speech feature extraction is considered.
In this paper a method is proposed to provide enhancement of the intelligibilty of speech which has been contaminated by additive noise. The method is based on the assumption that intelligibility enhancement can be achieved by making use of information directly related to speech intelligibility. A structure which provides the means to implement the method is presented and a discussion of some preliminary results obtained with the method is included.
A high performance load torque observer for impact drop compensation of a motor drive system has been realized by using a digital signal processor (DSP). This observer unit realizes high speed response and low noise characteristics by the DSP's software. The software is constructed with three functions; An improved load torque observer for high speed estimation, a moving average filter for eliminating torque pulsation component caused by inequality in firing angles of converter arms, and a variable filter for reducing noise contained in the estimated load torque. Use of this observer unit with a motor drive system offers high stability and impact drop compensation effect.
Since the variations of rotational speeds can provide significant information in diagnosing a rotating machine. Substantial effort has been expended in attempts to find the detailed variations of speeds. Speed variations have been estimated with varying degrees of success. A portion of the effort has been motivated by active research and development of monitoring systems for turbines and oil-drilling rods. The viability to accurately estimate small variations using noise contaminated records can be done by constructing a modified digital demodulation scheme.
An study on command transmission security in complex systems is presented. The main objective is to garantee error rates under the limits recommended by the international organizations, keeping the communication channel occupation as low as possible. Two methods increasing the security are analysed and compared: the redundant bits addition and the command message repetition. Using simulation techniques, the results are applied to the command transmission in the power control centers.
This paper presents history and recent developments in the field of systolic processing and its application to the domain of digital signal processing. The first part of the paper discusses both theoretical and practical milestones in systolic processing - from cellular automata to the Warp machine. In the second part various application examples are given and discussed.
An algorithm for filtering noisy step-like signals is proposed. This algorithm is based on the assumption of Gaussian contamination. In this procedure data within a moving window is divided into two almost equal clusters and a hypothesis tests (F-test) for differences in the means between two such clusters. Histogram analysis and/or our a priori knowledge about the number of discrete amplitudes in the ideal noise free signal provide information that is used to filter the signal further and produce a clean signal with the desired number of discrete amplitudes (levels). As an illustration the method is tested by simulation.
The new method for analysis of stationary (ergodic) random data in discrete time-series format using the concept of a correlation technique is proposed here. The validity of the presented method and its performance has been proved on the example of commutation study direct current (DC) machine.
Quantization noise due to interface devices of finite wordlength and the effects of this noise on the performance of a control system are analyzed. A digital filtering scheme for minimizing the effect of quantization noise along the derivative path of the control algorithm is developed and implemented. Results obtained by digital simulation are verified experimentally.
Various applications of the TMS32010 as an audio signal processing system are discussed. Although this chip is a first generation part, it contains most of the architecture necessary for real time DSP in the audio frequency range. A number of experiments that can be performed with a minimum of hardware are described. Included are spectrum shifting, digital filters, spectrum analysis, and adaptive filters. This material should be useful to independent research groups and universities embarking on new DSP programs.
In this paper, we compare abilities of two kinds of digital signal processor ( DSP ), M88764 and DSSP1. At first, we implement FIR digital filter and IIR digital filter in using each processor. Next, we implement a discrete Fourier transform by prime factor algorithm in each processor. The numbers of machine cycles and program steps to implement the algorithm in these processing are compared. Moreover, we make it clear how architecture of DSP is suitable for each digital signal processing.
A balance platform was used to collect postural stability data from 60 children, approximately half of whom have been diagnosed with cerebral palsy. The data was examined with respect to its frequency content, resulting in an improved strategy for frequency estimation. With a reliable assessment of the frequency domain characteristics, the signal stationarity could then be examined. Significant differences in signal stationarity were observed when the epoch length was changed, as well as between the normal and cerebral palsy populations.
Conventional parameters monitored in intensive care units are usually late indicators of neurologic deteriorations; therefore, clinical examination remains the cornerstone of assessing neurologic function and guiding treatment in head injury. In recent years considerable interest has developed in the use of evoked potentials as a possible adjunctive monitoring technique. This paper describes an automated on-line data acquisition system and the signal processing techniques used to acquire and interpret the data.
Biomechanics research at the Medical College of Wisconsin is directed to the determination of the mechanisms of head and spine injury and the evaluation of surgical treatments for these injuries. This work involves mechanical testing of components of the spine (disks, vertebral bodies, and ligaments) as well as testing of composite spines and in situ evaluation of intact human cadavers (1,3). Other studies utilize experimental animals to measure neurologic and physiologic effects due to injury producing loads and accelerations (2). An integrated system has been developed to facilitate the acquisition and analysis of the diverse types of data from these experiments.
A low-cost data acquisition system based on Z80 microprocessor has been developed to monitor and record the engine running hours of terminal tractors in port cargo handling operations. The engine hours data recorded on a large fleet of tractors will be retrieved electronically and later transferred to a personal computer in the office ready to be processed by standard database software package for maintenance planning and scheduling. The implementation of this system thus reduces manpower and time, and climates human error in data handling.
Recent process control systems show a tendency to grow increasingly in scale and become more complicated with the progress of instrumentation system. To achieve economical data transmission an: acquisition of an enormous amount of information, "Hybrid Local Area Network" has been developed. This LAN can be connected with a computer or a control device and has compaibility with sensors and actuators in terminal station. Normally, the data transmission and acquisition is performed by the roll-call poling method. In emergency, random access signals are transmitted by the contention method. This method is a highly flexible multiplex transmission which will be met the microcomputer based distributed process control system.
A microcomputer controlled approach velocity (differential distance) sensor and measurement system based on ultrasonic principle is presented. The measurement system is independent of any variation in atmospheric conditions, provided the target and sensor are subjected to the same atmospheric conditions. The system avoids need of recalibration, which is common in other methods implemented using compensating elements. It has got simplicity of design with low cost components and high potential accuracy. The sensor is compact and can be mounted on a robot gripper or on a CNC machine. The proposed system is suitable for usually encountered gripper velocities, since response time of the system is of the order of 25 ms for a target to gripper distance of 3 meters.
A new variable reluctance transducer is described. Two static coils constitute two branches of a Wheastone bridge. The magnetic circuit of each one of the windings is closed through two gear wheels attached to the shaft. These are related to each other in such a way as to leave two air gaps, each of these in front of one of the coils in the stator assambly. These gaps change with the torque.
This paper reports on the hardware and the programming aspects of an intelligent data acquisition and load forecasting system that has been implemented on a desktop microcomputer. The objective was to develop a low cost and reliable system that would collect forecasted weather data, real-time electric utility load data, archive them, and issue an electric utility load forecast in 1-hour, 6-hour and upto 24-hour increments within a midnight-to-midnight time frame. Data are collected, over commercial telephone lines, from remote locations (often hundreds of miles apart), filtered and then processed. The archived data are used to form monthly summaries of hourly electric utility load (MW) and weather conditions in the area. A set of pre-selected rules are then applied on this database to develop the desired load forecast. All this work is done in a totally automated fashion, i.e., without any human intervention. The data acquisition and load forecasting system is based on an AT&T 3B2/300 UNIX based desktop microcomputer. The 3B2 serves as the "heart" of the system and performs the functions of data collection, processing, archiving, load forecasting and display. It is a multi-tasking, multi-user machine and at it's present configuration can support four users and a "super user", or system manager.
A discrete position encoder is an inexpensive means for sensing the angular position of a rotating device. Often a system with higher performance can be achieved if the angular velocity is known in addition to the position. Typically, the output of a discrete position encoder is two square wave signals in quadrature. This paper investigates various methods for processing these signals to observe the velocity in real time. High performance observers based on Taylor series expansions, backward difference expansions, and least square curve fits are developed. The accuracy of the different observers are analyzed by simulations for systems with time measurement truncation and imperfect encoders. The least square curve fit based observers are the most tolerant observers investigated due to the inherent low pass filtering.
On-Line Fiber Optic Infrared Analyzers monitor the chemical composition of processed liquids in chemical plants. Signal processing algorithms utilize infrared spectral values and their derivatives to determine chemical composition. This paper will review some of the problems in developing these algorithms for different chemical applications.
We have realized a flat coil on a printed circuit with epoxy resin support. This way we easily obtained single or double faced flat coils. By incorporating such a coil into an oscillator we have obtained an inexpensive manufacturing proximity sensor. It is an eddy-current transducer which uses the principle of impedance variation. An eddy-current is induced into a conductive metal target and monitored by the proximity detector. The study of the transducer frequency response has permitted us to develop some applications like the detection of an edge of sheet metal or the accurate location of a circular aperture in an iron sheet. From the main sensor properties we have developed three different kinds of devices.
This paper deals with the hardware design of a real-time image preprocessor to service the operation of the DIAS system. DIAS is an autonomous, multiprocessor picture information system consisting of four hierarchical levels of operation (Master Planning, Distributed Service Control, Execution and Backend). The pre-processor proposed here receives parallel image data from the environment and provides to DIAS Master-Planner-Processor a number of image features (parameters) in order to service the decision making about the input image. These image features are: size (number of pixels); type (binary, or grey level picture); total average intensity of the picture; uniformity of the picture; number of informative square areas and their size; geographical locations of the informative picture regions and their grey level averages.
This paper describes a low cost system that simul-taneously provides a data and a voice channel over a standard general service telephone network (GSTN) 2-wire connection. Time domain harmonic scaling (TDHS)  is used to frequency compress the voice signal. Data rates up to 1200 bits/s full-duplex have been achieved using a low complexity quadrature amplitude modulation (QAM) algorithm. A real-time implementation has been built using three TMS32010 DSP chips for full-duplex operation.
A knowledge-based system is presented that is oriented toward partially occluded 2-D workpiece recognition in TV camera images. The generalized Hough transform is employed to extract elementary edge patterns. Intrinsic and relational information regarding elementary patterns is computed and then stored inside a net of frames. A similar net of frames is employed for workpiece model representation, for an easy matching with the previous net. A set of production rules provide the heuristics to find hints for locating focus-of-attention regions, while other production rules specify modalities for applying a hypothesis-generation-and-test process. Experimental results on a set of 20 workpieces are reported.