Rhyme Test Words Research Articles

Distinctive feature analysis of errant responses to Diagnostic Rhyme Test words.

Ten subjects responded in a conventional multiple-choice format to two randomized lists of Diagnostic Rhyme Test (DRT) words in white noise at 3-dB signal-to-noise ratio. The DRT consists of 96 word-pairs whose initial phonemes differ by one feature: voice-unvoiced, nasal-oral, sustained-interrupted, sibilated-unsibilated, grave-acute, and compact-diffuse. List one consisted of DRT words counterbalanced for each feature and list two contained the contrasting feature. Six replicates were obtained for each subject. Errant responses were analyzed for DRT features as well as by a Chomsky-Halle feature system that permitted assigning numerical values that ranged from one to six to phonetic differences in DRT word-pairs. Results are consistent with previous findings for equality of percent correct across DRT word lists. However, DRT feature contrasts within word pairs were not equally difficult. For example, sibilated and unvoiced stimuli were significantly more difficult than their unsibilated or voiced counterpart, respectively. Also, there was no significant correlation between word error rate and the corresponding Chomsky-Halle feature value for differences in word-pairs. Thus, word difficulty appears to be unrelated to degree of feature difference in a word-pair, but rather is associated with relative phoneme audibility for the type of noise employed.

Analysis of yes–no and confidence-rating word recognition response formats

The effects of two response formats (yes–no and confidence-rating) on percent correct, d′, and response criterion were examined. Diagnostic Rhyme Test words were presented in white noise (3 and 6 dB signal to noise ratios) to eight normal hearing subjects at 70 and 90 dB SPL; three replicates were obtained. For the yes–no format, subjects checked ‘‘yes’’ or ‘‘no’’ that a single response option matched the stimulus. For the confidence-rating format, subjects checked 1 of 2 response options and indicated confidence, or lack thereof, that the response matched the stimulus by marking ‘‘yes’’ or ‘‘no.’’ Both formats yield signal detection theory (SDT) estimates of d′ and response criterion, but differ in categorizing responses as hits, misses, false alarms, and correct rejections. A significant difference was found between response formats for percent correct and response criterion but not for d′. The effects of signal to noise ratio were significant but parallel across response format for percent correct, d′, and response criterion. Results indicate that response format may significantly affect conventional measures such as percent correct but not SDT measures that separate sensitivity and response criterion. The use of SDT measures in word recognition procedures is supported.

Telephone communications with several commercial respirators.

Previous work showed that telephone communications while wearing military respirators degraded both word comprehension and recognition speed. In addition, electronic amplification of the speech diaphragm signal had shown no advantage to the extra hardware. This experiment was performed to test effects of different configurations of commercially available respirators on telephone communications accuracy and speed. Twelve pairs of subjects were separated into different rooms and communicated by telephone. Modified rhyme-test words were presented by computer to the speaker, who transmitted the word by telephone to the listener. During the first replication, subjects were given no instruction about telephone communications procedure. During the second replication subjects followed a communications protocol that instructed them when to move the telephone handset from their ears to their mouths. Results showed that the protocol uniformly improved communications accuracy without incurring any extra time penalty. Word comprehension was still twice as fast without a respirator as with a respirator. Accuracy with the protocol nearly equaled the no respirator control value for most respirators tested.

Performance of a Backpropagation Neural Network in Diagnostic Rhyme Test Word Recognition

This paper investigates isolated speaker- dependent word recognition of Diagnostic Rhyme Test words using a backpropagation neural network classifier. The performance of K-nearest neighbors and closest-class-mean classifiers are compared for several signal- to-noise ratios. The test and training data consisted of 40 frames of weighted eighth- order cepstral coefficients extracted from each word utterance. The neural network classifier correctly classified more than 92% of 2,400 testing examples not contained in the training data for the noise-free case. This performance was better than that of the K-nearest neighbor classifier, which was greater than 83%, and that of the closest class mean classifier, which was greater than 85%.

Priming the visual recognition of spoken words.

A preliminary investigation was conducted to understand the effects of word visibility and prime association factors on visual spoken word recognition in lipreading, using a related/ unrelated prime-target paradigm. Prime-target pairings were determined on the basis of paper-and-pencil word associations completed by 85 participants with normal hearing. Spoken targets included 60 single-syllable Modified Rhyme Test words, prerecorded on laser video disc. Participants included 20 individuals with normal hearing and at least average lipreading skill for sentence-length materials. In related prime-target pairings, more targets with a high prime association were identified than with a low prime association. In unrelated prime-target pairings, a larger number of more-visible than less-visible targets was correctly identified. Individual participant differences were not statistically significant. Results from the present study suggest implications for models of visual spoken word recognition.

Distinctive feature-based scoring of errant responses to rhyme test words

A distinctive feature-based system [W. A. Wicklegren, J. Acoust. Soc. Am. 39, 388–398 (1966)] for scoring errant responses to rhyme test words was developed. Words were presented in noise (0, 4, 8, and 12 dB signal-to-noise ratio) at four intensities (20, 40, 60, and 80 dB HL) to five normal hearing subjects. Errant responses were assigned values from 1 to 7, and categorized as mild, moderate, or severe, according to number of features missed. A basic computer program analyzed responses and yielded a total feature error score (TFES), the number of mild, moderate, and severe errors, and a transformed percentage correct score in rationalized arcsine units (rau) for each 50-item list. The TFES and rau data yielded parallel functions for signal-to-noise ratio and intensity. The correlation coefficient for TFES and rau was −0.965. An ANOVA for rau yielded three significant two-way interactions, the TFES only one. As conditions became difficult and percentage correct diminished, listeners tended to make more mild errors rather than severe. Results have implications for development of word recognition tests.

Differences in speech discrimination in the elderly as a function of type of competing noise: speech-babble or cafeteria.

Speech discrimination of place-loaded Modified Rhyme test words in speech-babble and cafeteria noise by elderly hearing-impaired subjects was measured. The stimuli were presented under the following conditions: (1) monotic: the signal was presented to the preferred ear; (2) low-frequency attenuated: the frequency band below 1,000 Hz was attenuated by 5, 10, and 15 dB relative to the high-frequency band; (3) dichotic: the frequency band above 1,000 Hz was presented to the preferred ear and the low-frequency band to the other ear; (4) diotic: the same signal was presented to both ears. Difference scores between the monotic baseline condition and the other listening modes were calculated and compared. Relationships between listening modes differed as a function of type of competing noise. Low-frequency attenuation improved discrimination in speech-babble but not in cafeteria noise. Dichotic-monotic and diotic-monotic superiority were seen with both noises.

Effects of Dichotic/Diotic versus Monotic Presentation on Speech Understanding in Noise in Elderly Hearing-Impaired Listeners

Hearing-impaired elderly subjects with mild to moderate, bilaterally symmetrical, sensorineural hearing losses received speech stimuli mixed with cafeteria and speech-babble noises in a variety of listening conditions. Open and closed set Rhyme test words and Speech Perception in Noise test sentences were presented at individually determined signal-to-noise ratios. the listening conditions included: (1) monotic against which all other conditions were compared; (2) low frequency attenuation of 5, 10, and 15 dB relative to the high frequencies; (3) dichotic, with the frequency band above 1000 Hz presented to the preferred ear and the band below 1000 Hz to the other ear; and (4) diotic, with identical speech-in-noise signals presented simultaneously to both ears. Results indicated: (1) regardless of type of stimulus, open or closed set presentation, or type of noise, most of the dichotic and/or diotic presentations were superior to monotic at 0.05 or 0.01 significance levels; (2) under most conditions, dichotic/diotic conditions were superior to the low frequency attenuated presentations; (3) subjects with abnormal Synthetic Sentence Inventory functions seemed to use dichotic/diotic input as well as those with normal Synthetic Sentence Inventory performance; (4) subjects with borderline hearing losses showed greater amounts of dichotic/diotic superiority than those with mild-to-moderate losses. Further research with wearable hearing aids is indicated.

Speech perception for various time constants of amplitude compression

Modified rhyme test words and CVC nonsense syllables were processed without or with background pink noise through a wideband amplitude compressor set for a 2.5:1 compression ratio and various attack (TA = 1 through 42 ms) and release (TR = 10 through 370 ms) times. These stimuli were presented to nine subjects with sensori-neural hearing impairment. The compression threshold was set 25 dB below the average maximal rms level of the vowels in the test words or CVC syllables. The compressed signal was amplified with individually adjusted frequency responses and delivered to the listeners via a circumaural earphone. The presentation of speech was always at the same, comfortable level for each individual listener. The signal compressed without noise was sometimes mixed with a masking pink noise at − 10, − 5, and 0 dB re speech. There were no significant differences in mean performances for different time constants, or between compressed and noncompressed stimuli. Performance slightly increased with compression, only when masking noise was added to the signal after processing (Ls − Ls ⩽ 5 dB) and the attack and release times were shorter than 3 and 90 ms, respectively. [Work supported by NIH NS 12946.]

Intelligibility of low‐ and high‐pass filtered speech set at 1900 Hz for hearing‐impaired subjects

Speech intelligibility for Modified Rhyme Test words was measured through a 1900‐Hz low‐pass and a 1900‐Hz high‐pass filter condition. The stimuli was presented monuaurally to twelve subjects with a moderately sloping sensori‐neural loss at a 20, 30, and 40 dB level above the subject's speech detection threshold. The speech discrimination scores were significantly better through the low‐pass filter condition over the high‐pass filter condition. No significant difference was observed as a function of sensation levels, or for the interaction of filter conditions and sensation levels.

Intelligibility Improvement of Analog Communication Systems Using an Amplitude Control Technique

Mobile satellite relay voice communications systems must provide high quality and reliable voice communications using relatively simple and inexpensive mobile equipment and easily achievable effective isotropic radiated power. High noise level reception will occur with analog voice communication systems operating over a power-limited link. At high noise levels, the low-level speech phonemes become masked by noise, which results in reduced speech intelligibility. In addition, a listener subjected over an extended period of time to this high noise level can become discomforted to the extent that his ability to understand what is being communicated is significantly reduced. An amplitude control technique has been employed for use with analog voice communication systems, which improves low-level phoneme reception and eliminates the received noise between words and syllables. Tests were conducted at the NASA Goddard Space Flight Center on a narrow-band frequency-modulation simplex voice communication channel employing the amplitude control technique. Presented for both the modified rhyme word tests and the phonetically balanced word tests are a series of graphical plots of the tests' score distribution, mean, and standard deviation as a function of received carrier-to-noise power density ratio. At low received carrier-to-noise power density ratios, a significant improvement in the intelligibility was obtained. A voice intelligibility improvement of more than 2 dB was obtained for the modified rhyme test words, and a voice intelligibility improvement in excess of 4 dB was obtained for the phonetically balanced word tests.

A study of pharyngeal speech.

Pharyngeal speech represents one of several types of alaryngeal speech; however, its use as a primary method of communication is rare. This report relates the principal findings of an intensive study of a 12-year-old girl with laryngeal papillomatosis who has used pharyngeal speech as an exclusive method of oral communication since age two. The unique physiologic mechanisms of pharyngeal speech are described and differentiated from other forms of alaryngeal speech. This girl’s reduced pharyngeal speech intelligibility for consonant and vowel rhyme-test words, her unfavorable phonation time and maximum phonation duration characteristics, her low average fundamental frequency, and her markedly hoarse pharyngeal voice quality all are distinct vocal liabilities. These findings lend strong support to the hypothesis that pharyngeal speech should not be regarded as a desirable or practical primary method of alaryngeal speech.

A study of buccal speech.

An intensive study of a normal-speaking subject, proficient in the use of buccal speech, was conducted. With respect to voice fundamental frequency variability, phonation time, and speaking rate his buccal speech characteristics compared favorably with those reported for excellent esophageal speakers. However, the reduced intelligibility of his buccal speech on rhyme-test words, the high average fundamental frequency of his buccal voice, and his conspicuous buccal gestures during speech represent distinct vocal liabilities.

Speech Intelligibility in a Stationary Multipath Channel

The intelligibility of speech transmitted through an acoustic channel such as the ocean is limited by multipath “time smearing.” The purpose of this study has been to obtain a quantitative measure of the effects of such time smearing on speech intelligibility. A linear, time-invariant channel was used as a model of the ocean. The impulse response of this channel was a sample of band-limited Gaussian noise. Using fast Fourier transform techniques, words of the modified rhyme test were convolved with (“smeared by”) this channel impulse response. The intelligibility of these “smeared” words was measured as a function of the impulse response duration T. Intelligibility was found to decrease monotonically as a function of T over the range T = 0 to T = 150 msec. Increasing T beyond 150 msec caused little change in intelligibility. With rhyme-test words used in a 6AFC test, average scores of 75%–80% were obtained for T between 150 and 665 msec.