Herbert J. Oyer

An international comparison of long‐term average speech spectra

The long‐term average speech spectrum (LTASS) and some dynamic characteristics of speech were determined for 12 languages: English (several dialects), Swedish, Danish, German, French (Canadian), Japanese, Cantonese, Mandarin, Russian, Welsh, Singhalese, and Vietnamese. The LTASS only was also measured for Arabic. Speech samples (18) were recorded, using standardized equipment and procedures, in 15 localities for (usually) ten male and ten female talkers. All analyses were conducted at the National Acoustic Laboratories, Sydney. The LTASS was similar for all languages although there were many statistically significant differences. Such differences were small and not always consistent for male and female samples of the same language. For one‐third octave bands of speech, the maximum short‐term rms level was 10 dB above the maximum long‐term rms level, consistent across languages and frequency. A ‘‘universal’’ LTASS is suggested as being applicable, across languages, for many purposes including use in hearing aid prescription procedures and in the Articulation Index.

Experiment on Voice Identification

A two‐year experiment on voice identification through visual inspection of spectrograms was performed with the twofold goal of checking Kerstas claims in this matter [Nature 196, 1253–1257 (1962)] and testing models including variables related to forensic tasks. The 250 speakers used in this experiment were randomly selected from a homogeneous population of 25 000 males speaking general American English, all students at Michigan State University. A total of 34 996 experimental trials of identification were performed by 29 trained examiners. Each trial involved up to 40 known voices, in various conditions: with closed and open trials, contemporary and noncontemporary spectrograms, nine or six clue words spoken in isolation, in a fixed context and in a random context, etc. The examiners were forced to reach a positive decision (identification or elimination) in each instance, taking an average time of 15 minutes. Their decisions were based solely on inspection of spectrograms; listening to the identificati...

Ratings of Vocal Pleasantness and the Aging Process

In response to the questions raised at the outset, the following summary statements can be made: (1) Listeners were reliable in making pleasantness judgments. All listening groups were reliable, but the two older listening age-groups had the highest correlation coefficients. (2) Speakers were judged as less pleasant as their age levels increased. (3) Young listeners made judgments nearer the unpleasant end of the scale than did the older listeners. (4) Male speakers were judged as more pleasant than were female speakers. (5) Female listeners rated speakers as being more pleasant than did the male listeners.

Relative Intelligibility of Speech Recorded Simultaneously at the Ear and Mouth

Monosyllabic words recorded at the lips and left ears of six speakers were fed to the headsets of 24 trained listeners at −12, −15, and −18 S/N ratios. Although the trend for intelligibility scores throughout the test is in the same direction for signals of both origins, decreasing S/N ratio is more destructive to the speech picked up at the lips.

Latest Developments in Voice Identification

A 2‐yr voice‐identification experiment was completed at Michigan State University in December, 1971. A total of 34 996 experimental trials of identification were performed by 29 trained examiners, involving a variety of conditions. Those experimental conditions correlated with forensic models (open trials fixed and random contexts, noncontemporary spectrograms) yielded an error of approximately 6% of false identifications and 12% of false eliminations. Examiners were forced to reach a positive decision in each experimental trial. They judged approximately 60% of their wrong answers and 20% of their right answers as “uncertain.” This suggests that if the examiners had been able to express no opinion when in doubt, only 74% out of the total number of tasks would have had a positive answer with approximately 2% errors of false identifications and 5% errors of false eliminations. The experience obtained in both laboratory and field work, during more than 3 yr, convinced the authors that voiceidentification te...

Temporal speech characteristics and vocal pleasantness in the aging voice

Previous work by the authors emphasized temporal characteristics of speech as critical cues in age estimation. The present study assessed the role of the same temporal cues in listener perception of vocal pleasantness (PL). Twenty‐four speakers (12 male), ages 41–82 years, read the passage “My Grandfather” and prolonged /aeiou/. With the first two sentences of “My Grandfather” as the stimulus, 22 college‐age naive listeners (11 male) judged speaker PL on a five point scale. The times to initiate and to terminate /aeiou/, the time to articulate /maɪ gnaendfðɚ/, reading rate, pause time within the passage, and speaker sex were regressed against PL. The single best predictor for PL was reading rate (R‐square = 0.48) and the best significant model (12 variables) for PL yielded R‐square = 0.77. Similar to age estimation, the perception of PL appeared to be strongly influenced by temporal characteristics of speech. While the contributions of individual variables differ, the results suggest that estimations of ag...

An Experiment on Voice Identification by Visual Inspection of Spectrograms

An experiment aimed to test Kerstas claims on the reliability of speaker identification by visual inspection of spectrograms is being performed under a contract with the Michigan State Police, sponsored by the Federal Department of Justice. Thirty‐eight examiners were selected, trained during approximately three weeks, and then employed to perform a total of approximately 32 000 trials of identification. Two hundred and fifty speakers, taken randomly from a homogeneous male student population of 25 000 twice recorded, one month apart, five utterances of nine clue words, spoken in three different conditions. Approximately 54 000 spectrograms (broad‐band, 4‐kHz, Voiceprint spectrograph) were obtained from these recordings and used in the trials of identification performed by the examiners. One‐third of these trials were closed, one‐third were open, including a matching spectrogram, and one‐third were open trials not including a matching spectrogram. Several levels were tested in this experiment, involving ...

The differential performance of older speakers in initiating and terminating phonation as a function of health status

The purpose of this study was to determine the utility of vowel initiation tune (VIT) and vowel termination time (VTT) in discrimination between healthy and unhealthy elderly speakers. Sixty‐three adults age 60–94 years sustained /a, e, i, o, u/ for at least 1 s. From audio recordings of these productions means and standard deviations of VIT and VTT were derived. A MANOVA based on speaker gender (36 females, 27 males) and speaker health status (42 unhealthy, 21 healthy) yielded a significant difference in performance in VIT as a function of health status but for gender on the interaction. A subsequent linear discriminant analysis using mean VIT correctly classified subjects in the unhealthy group with 33.3% accuracy (3 times greater than chance levels) and subjects in the healthy group with 95.2% accuracy. These results support the hypothesis that the degree of coordination between respiration and phonation reflects the health status of the older adult.

Intra‐speaker analysis of the nasal consonant [m]

The acoustic spectrum of the nasal consonant [m] was analyzed from utterances of subjects as they produced it (1) in monosyllables, (2) in the reading of a paragraph, and (3) in attempts to disguise their voices while speaking a sentence. Findings indicate that although there is considerable intra‐subject variability in the overall spectrum of the [m] consonant, the first formant remains remarkably constant. LPC analysis was employed in extracting acoustic characteristics. The results of this study have implications for speaker identification.

Intraspeaker variability on nasal consonant [m]

Our previous study determined that intraspeaker variability is small for the first formant of the nasal/m/when read in a carrier sentence and larger when read in a paragraph. This study investigates factors responsible for increase in intraspeaker variability. Syllables [mi], [ma], and [mu] were each read ten times and recorded (1) in a carrier sentence with a slow rate, (2) in a carrier sentence with rapid rate, and (3) in a paragraph with normal reading rate. First formant frequency contours were compared to determine effects of context and rate on intraspeaker variability. Findings suggest that intraspeaker variability of the nasal /m/ is closely related to context, but rate of utterance is somewhat unimportant. The implication is that, although the /m/ is produced within a vocal tract that remains relatively constant, it is still subject to some variation due to the effects of coarticulation. A comprehensive study of the sounds subject to the least coarticulatory effects may prove to be important in s...