Probability Of Correct Recognition Research Articles

How do humans process and recognize speech?

Until the performance of automatic speech recognition (ASR) hardware surpasses human performance in accuracy and robustness, we stand to gain by understanding the basic principles behind human speech recognition (HSR). This problem was studied exhaustively at Bell Labs between the years of 1918 and 1950 by Harvey Fletcher and his colleagues. The motivation for these studies was to quantify the quality of speech sounds in the telephone plant to both improve speech and preference. To do this he and his group studied the effects of filtering and noise on speech recognition accuracy for nonsense consonant-vowel-consonant (CVC) syllables, words, and sentences. Fletcher used the term as the probability of correct recognition for nonsense sounds, and intelligibility as the probability of correction recognition for words (sounds having meaning). In 1919, Fletcher found a way to transform data for filtered speech into an additive density function D(f) and found a formula that accurately predicts the average articulation. The area under D(S) is called the articulation index. Fletcher then went on to find relationships between the recognition errors for the nonsense speech sounds, words, and sentences. This work has recently been reviewed and partially replicated by Boothroyd and by Bronkhorst, et al. (1980). Taken as a whole, these studies tell us a great deal about how humans process and recognize speech sounds. >

Myocardial function in early hours after coronary artery bypass grafting: Comparison of two cardioplegic methods

The theoretical advantages of retrograde blood cardioplegia combined with anterograde blood cardioplegia and warm reperfusion before aortic unclamping during coronary surgery were evaluated in 41 patients (group 2). The early postoperative myocardial function of this group was compared with that of 55 patients (group 1) in whom cold crystalloid cardioplegia was administered. The following variables were measured and analyzed by multivariate statistical analysis: heart rate, left atrial pressure, systemic arterial pressure, cardiac index, left ventricular stroke work index, ventricular function, oxygen delivery, hemoglobin, partial oxygen pressure in mixed venous blood, arteriovenous oxygen difference, carbon dioxide production per square meter, and cardiac isoenzyme of creatine-kinase. The myocardial function improved progressively and cardiac enzymatic release was low for both groups 9 hours after admission to the intensive care unit. However, group 2 had significantly higher oxygen delivery, carbon dioxide production per square meter, cardiac index, left ventricular stroke work index, and ventricular function and significantly lower left atrial pressure and mean systemic arterial pressure than that of group 1. The beat separation of group 2 from group 1 occurred at the ninth hour, with a probability of correct recognition of 92.1%.

Variable selection for the classification of postoperative cardiac patients

A set of 13 extensively used hemodynamic, ventilatory and gas analysis variables are measured (on-line or off-line) on 200 patients in an intensive care unit (ICU) during the 6 h immediately following cardiac surgery. Since the existence of two well-separated classes of low- and high-risk patients has been previously shown the divergence criterion is then used to identify those variables which, at three equidistant observation times, possess the greatest separation power. Such variables always include the cardiac index (CI). representative of cardiac performance, and two indices related to respiratory efficiency and metabolic rate, i.e. the carbon dioxide production index (VCO 2I) and the arterio-venous oxygen difference (avO 2D). The Fisher linear classifier, utilizing these three features, is then tested by using the rotation method. The results show good performance of the linear classifier, which exhibits a probability of correct recognition always greater than 87%, thus suggesting the possibility of obtaining interesting improvements by means of more sophisticated classifiers.

Predicting Memory for Components of TV Commercials from EEG

Subjects watched television while EEG was being recorded and later completed a series of recognition tests based on component parts of the commercials they had seen. Memory correlated significantly with changes in the electrical patterns that occurred during viewing. The probability of correct recognition was enhanced when alpha blocking continued for a longer period of time and when hemispheric laterality shifted to the right during the onset of a commercial component and then to the left during the following seconds.

Three-dimensional object recognition from multiple views

We discuss an algorithm that enables a machine-vision system to recognize a three-dimensional object from a series of multiple views. The algorithm uses pseudoreflection tomography to map N images of the object into a single signature image that incorporates information from all N views. Rotation-, scale-, and brightness-invariant features are extracted from the signature image that permits robust recognition of the object. The signature image enjoys other invariants as well, such as invariance with respect to both translation of the yaw (i.e., y) axis and slight vertical shifts of the object along the yaw axis. Rotation-invariant features are extracted by using the circular-harmonic functions originally introduced by Hsu [ Appl. Opt.21, 4012 ( 1982)]. Scale invariance is obtained by using Mellin transforms. If only partial-view data are available, the probability of correct recognition is obviously decreased but can to some extent be ameliorated—at least in a low-noise environment—by using similarity measures between test and reference objects based on gray-level coincidence statistics.

Cues for lexical stress and recognition of polysyllabic words, synthesized from diphones and presented in isolation

Knowing the position of lexical stress in a polysyllabic word may considerably limit the number of lexically possible responses. Thus perceptual cues may contribute to a higher probability of correct recognition and faster recognition of words in the perception of speech of less than optimal quality. Probability correct was assessed for 30 three‐syllable Dutch words, synthesized from diphones. All words were of CVCVCVC phonemic structure, containing no consonant clusters. Ten words each had initial, medial, or final lexical stress. Stress position was either not cued at all, cued by durational cues only, cued by a conspicious pitch movement, or cued by both durational and pitch cues. Each word in each condition was responded to by five listeners. Scores correct, averaged over stress position, were 66% with no stress cues, 74% with durational cues, 81% with pitch cues, and 81% with durational and pitch cues combined. The contribution of stress cues was strongest for words with initial stress (from 54%–76%), and minimal for words with final stress (from 80%–88%). To investigate the effect of stress cues on the rapidity with which listeners tune in on the correct response, a gating‐paradigm‐type of experiment was performed on the same material. However, in this task errors against stress position were much more frequent, and positive effects of stress position were small.

Memory for at least one of two items: Test and failure of several theories of spacing effects

Spacing between two attempts to learn one item greatly increases the probability that it will be remembered later. We first note that a number of theories that explain this phenomenon also predict that spacing should improve the probability of remembering at least one of two different items each studied once. We then report results of two experiments designed to test this prediction. Subjects studied a series of 300 words. After a short rest they were tested for recognition, or—in the second experiment—free recall and then recognition. As usual a substantial spacing effect was observed for single words presented twice. In contrast the probability of correct recognition or recall of at least one word of a pair of two different words, each presented once, did not depend on their spacing in the study series.

Image Quality and Face Recognition on a Television Display

Subjects were asked to match faces presented singly on a television display with one of 35 faces presented on a photographic display. The probability of correct recognition and the time required to recognize the single face were related to the quality of the television image. Image quality was varied by changing the square-wave response of the television system and the video signal-to-noise voltage ratio. A derived unitary metric of image quality, the square-wave modulation transfer function area, MTFASQ, was shown to correlate highly with both measures of observer performance. The utility of this metric for systems design is discussed.

The effects of TV camera field of view and size of targets upon air-to-ground target recognition.

A simulation experiment was performed to determine the effects of TV camera lens field of view and target size upon air-to-ground target recognition via closed-circuit television. Measures of performance were probability of correct target recognition, range of correct recognition, and proportion of errors committed. As the field of view decreased, (1) probability of correct recognition decreased (P < .01), (2) mean range of correct recognition increased (P < .01), (3) errors of commission did not vary (P > .05), and (4) errors of omission increased (P < .01). As target size decreased, (1) probability of correct recognition decreased (P < .01), (2) mean range of correct recognition decreased (P < .01), (3) errors of commission increased (P < .01), and (4) errors of omission increased (P < .01). The increase in mean recognition range with decreases in field of view was greater for large targets than for small targets (P < .05). The field of view had little or no effect upon errors of commission for large targets; however, for small targets, the smaller the field the greater the proportion of errors of commission (P < .01).

Recognition of familiar characters under an unfamiliar transformation.

Pattern recognition by human beings and infra human organisms is remarkable chiefly because it can take place with speed and with a relatively low probability of error even when the incoming information has been subjected to a wide variety of transformations and extreme degradation. Recent attempts to build automata to recognize patterns have served to increase, rather than diminish respect for this capacity of living systems. Almost without exception, the existing, commercially available automata recognize patcerns only under conditions so restricted that a living organism would soon perish were it necessary for it so to limit the class of inputs co which it could respond with reasonable accuracy. Commercial number reading machines, for example, usually require chat characters come from a specially designed type fonc, that they be of a certain size, and chat they occur only on certain portions of the document to be read. The human reader does not operate under such restrictions. H e can read numbers from many type fonts, even from fonts which he has never before seen; nor does he have to be told the name of the fonc in order to be able to read it. The characters to be read do not have to be in one particular part of the page, nor do they need to be of one particular size. This human facility in pattern recognition has raised two sorts of problems. First it has led some to overstate the capabilities of the human pattern recognizer. W e do not recognize patterns with equal ease under all the rigid transformations. Recognition thresholds for letters presented upside down are measurably higher than for upright letters, though some discussions of human pattern recognition imply that it is invariant under rotation. The other problem is that the human being recognizes familiar characters so rapidly and so accurately under most conditions that it is difficult to obtain stable performance estimates which are sensitive to experimental manipulation. Tachistoscopic exposure has been the favored method for keeping observers from making close to 100% correct identifications from a known and familiar set of characters, but this is an awkward technique for obtaining a legibility measure on a new type font, for example. Other techniques have involved performing some transformation on the characters repeatedly until probability of correct recognition reaches some lower limit. The characters can be made smaller and smaller, for example, the technique used in the ordinary Snellen chart to measure visual acuity. Studies concerned primarily with transmission of visual information over communication

Recognition of forms against a complex background.

Although many visual tasks require the observer to search for and attempt to recognize one of a particular class of objects of interest, the problem has received little systematic investigation. Basic visibility data, based upon known shapes presented in known positions, certainly do not apply. In this experiment, observers were repeatedly presented with a background array of curvilinear forms, among which a “critical target” was present 50% of the time. The critical targets consisted of one or another of five examples of six geometrical forms, all with straight sides and sharp corners, any one of which was equally likely to appear in any part of the visual field. Probability of correct recognition was determined as a function of the number of background forms, viewing time, visual angle, and form density. Forms were selected on the basis of a preliminary experiment in which basic recognition data were gathered on 240 different forms.