Spondee Words Research Articles

Message Reception and Message Uncertainty

A basic principle of language engineering is that the accuracy of message reception in noise is determined by the number of alternative messages that might be communicated. However, it is not clear whether the number of alternative messages available for presentation or the number of possible message-responses is the critical determinant of performance. We attempted to resolve this question in experiments in which trained listeners first memorized various sets of n spondee words (n varied from 2 to 64). Words, chosen at random from the set of n words, were then read, one at a time, over a noisy communications channel. Shortly after the word was read, the listener was instructed that the presented word was one of k possible words, where k⩽n. Accuracy of message-reception was found to be practically independent of the size of the message class (n), for a constant number of possible response alternatives (k). Furthermore, the number of necessary response distinctions—rather than response-uncertainty per se—appears to be the critical determinant of accuracy of communications.

Some Relations between Normal Hearing for Pure Tones and for Speech

The present American standards for normal hearing for audiometric purposes imply that the difference between audiometric reference zero for a 1000-cps pure tone and audiometric reference zero for spondee words is 6 db (i.e., approximately 16 db SPL for a 1000-cps tone and 22 db SPL for spondee words). An accumulation of evidence from several sources suggests, however, that the difference between the two thresholds is considerably larger than 6 db. The present study was designed to assess the effects of sex, ear, order of test presentation, method of threshold measurement, and degree of prior familiarity with spondee vocabulary on the difference between the threshold for a 1000-cps tone and the threshold for spondee words, as measured with W-1 recordings. Subjects were 126 young adult college students. Results indicate that the difference varies from 12 to 14 db, depending slightly on degree of prior familiarity with the spondee words, and on the method used to measure the pure tone threshold. The average speech threshold SPL in the present group agreed closely with the current standard reference level of 22 db, while the average 1000-cps pure tone threshold SPL was 6 to 8 db lower than the current standard reference level of 16 db. [This research was supported by the U. S. Air Force under Contract No. AF 41 (657)-185, monitored by the School of Aviation Medicine, USAF, Randolph Air Base, Texas.]

Masking of English Words by Prolonged Vowel Sounds

One-hundred and ten monosyllabic words selected from the Thorndike list of 1000 most frequently occurring words in English to represent equally each of 10 vowels were presented to 300 American listeners in an articulation test. Also tested were 72 spondee words, half selected from those in use in audiological tests and half from those in frequent use in air traffic control, further to represent the same vowel sounds. Masking of the stimuli was accomplished by recording separately each of nine prolonged vowels intoned by a trio of male voices. Results indicate that vowels of equal sound pressure levels differ considerably in masking effectiveness, that words containing specific vowel are not masked optimally by the same vowel, and that spondees are masked by prolonged vowels in the same rank as are the monosyllables. Prolonged vowel sounds with relatively high concentrations of energy between 700 and 1000 cps are most effective as masking agents. Rank order correlation of observed masking effectiveness with masking effectiveness predicted by the Strasberg method is 0.52, the Beranek method is 0.59, and by the Pickett-Kryter method, 0.69. Some observations on resistance of words to masking are made in relation to phonemic transition areas within words.

Confidence Ratings, Message-Reception, and the Receiver Operator Characteristic

Can an observer rate the accuracy of his message reception? Apparently, yes. Listeners were presented with spondee words against a white noise background. For each word presented, the listener recorded his received message. Immediately thereafter, he assigned a confidence rating with respect to his accuracy of message reception. We find that the assignment of confidence ratings does not interfere with the accuracy of message reception. More important, the relative accuracy of message reception is directly related to the confidence rating. This relationship is relatively invariant over a range of speech-to-noise ratios. It may be noted that the rating procedure directly yields the entire trading relation between correct confirmations and false alarms—the receiver operating characteristic (ROC). The form of the ROC curves obtained by the rating procedure is independent of average word difficulty and is similar to that obtained by successive experiments with fixed acceptance criteria.

Masking of English Words by Prolonged Vowel Sounds

One-hundred and ten monosyllabic words selected from the Thorndike list of 1000 most frequently occurring words in English to represent equally each of 10 vowels were presented to 300 American listeners in an articulation test. Also tested were 72 spondee words, half selected from those in use in audiological tests and half from those in frequent use in air traffic control, further to represent the same vowel sounds. Masking of the stimuli was accomplished by separately recording each of nine prolonged vowels intoned by a trio of male voices. Results indicate that vowels of equal sound pressure levels differ considerably in masking effectiveness, that words containing a specific vowel are not masked optimally by the same vowel, and that spondees are masked by prolonged vowels in the same rank order as are the monosyllables. Prolonged vowel sounds with relatively high concentrations of energy between 700 and 1000 cps are most effective as masking agents. Rank order correlation of observed masking effectiveness with masking effectiveness predicted by the Strassberg method is 0.52, the Beranek method is 0.59, and by the Pickett-Kryter method, 0.69. Some observations on resistance of words to masking are made in relation to phonemic transition areas with words. [Done in connection with contract AF 19(604)-1577, monitored by the Operational Applications Laboratory, Air Research and Development Command, Bolling Air Force Base, Washington 25, D.C.]

Continuous Analysis Speech Band-Width Compression System

An investigative program conducted at Melpar under U. S. Air Force sponsorship has resulted in the development of a speech band-width compression system capable of transmitting speech within a total band width of 150 cps (as compared with 3000 cps for speech transmission by conventional means). This system consists of an analyzer which extracts from the acoustical speech signal six vocal analog control signals, a suitable transmission link, and a synthesizer which employs the control signals to reproduce the essential features of the original speech. The vocal analog control signals vary slowly in time because of natural physical constraints, and thus require considerable less transmission band width than the original acoustical signal. Articulation tests conducted on the system have resulted in scores of 67% for PB (phonetically balanced) word lists and 88% on Spondee word lists. A breadboard model of this system will be described and tape recordings of processed speech will be presented. [This work was supported by Contract AF33 (600)-3.]

Assessment of Impaired Hearing by Articulation Scores

The phonetically balanced “PB” lists are an effective tool for measuring the ability of impaired ears to hear correctly words of varying degrees of difficulty. If we know the position of a patient's PB articulation curve on the scales of intensity and of discrimination, we can estimate the percentage of ordinary words that he can hear adequately when speech is (1) faint, (2) average, and (3) loud. The average of these three scores constitutes our Social Adequacy Index for hearing. The location of the foot of a patient's articulation curve on the intensity scale can be most quickly and accurately determined by threshold tests once we know the average articulation score that corresponds to the particular threshold test employed. The equivalent PB articulation scores have been determined experimentally for Test No. 9 (spondee words), Intelligibility of Connected Discourse, Test 4C (2-digit numbers), and Test No. 12 (simple sentences).