Trained Listeners Research Articles

Acoustic measures of dysphonic severity across and within voice types.

The purpose of the study was to explore an interaction between pathologic voice type and the acoustic prediction of dysphonic severity. One hundred and two phonatory samples, representing a wide range of laryngeal conditions, were categorized by listeners into three voice types: breathy, rough, and hoarse. A second group of trained listeners rated the severity of the samples on a 7-point scale. Twenty-five frequency- and time-domain measures were used to predict perceptual severity. Multiple regression analyses showed that the most useful measure for the prediction of severity across voice types was noise-to-harmonic ratio (NHR). The severity of individual voice types was predicted with differential results on the basis of both short- and long-term measures of perturbation.

Language and dialect identification by syllabic spectral features

A speaker-based automatic language identification system has been studied to discriminate both languages and dialects. Syllabic spectral features are extracted automatically by an artificial neural network. The system makes no use of phonetic segmentation for recognition or training. High performance has been obtained for monologue telephone recordings. The system has been applied to several different databases to study the discrimination possible among different languages, dialects, and accents with such limited phonetic–acoustic information. The system performance is also related to individual speaker characteristics and the size of reference population for each language, and the length of testing samples. This study will report experimental results related to performance and data environments. It will also report an attempt by others to compare recognition performance of trained or expert human listeners with automatic recognition results.

Accuracy of same/different judgments of sequences of complex tones differing in tonal order under various levels of fundamental frequency range, listener training, and type of standard sequence

Same/different judgments of sequences of complex tones differing in tonal order were obtained so that the relationship of performance to the range among the sequential tones’ fundamental frequencies could be measured. Frequency range was defined as the difference between the mean of the two higher fundamental frequencies and the mean of the two lower fundamental frequencies which alternated with them; strengths and weaknesses of this measure are discussed. Data were taken from naive and trained listeners using fixed and random standards. Performance declined as frequency range increased. The performance of listeners trained with a variable standard and in a single condition of the study improved only at the narrowest frequency range; with a fixed standard, this training yielded improvements at a wider frequency range. The performance of listeners trained with a variable standard and in all the conditions improved at the narrow and moderate frequency ranges; with a fixed standard, this training yielded the highest levels of performance measured in each frequency range. The results are discussed in terms of subdivided temporal compounds, auditory attention, and memory for sequences.

Discriminating coherence in spectro-temporal patterns

This study examined the ability of trained listeners to discriminate coherent components in randomly varying spectral patterns. In each observation interval, the listener was presented with a sequence of bursts of multitone complexes having a fixed number of tones (m) in each burst. In the standard interval, the frequency of each tone in every burst was chosen randomly between 200 and 5000 Hz. In the signal interval, the frequencies of n tones were repeated throughout the burst sequence while the remaining m-n tones were chosen at random. The n tones were coherent in the sense that they were perceived as "sticking together" to form a pattern. The listener's task was to discriminate which burst sequence contained the n components. The results indicated that discrimination improved with increasing n/m, with increasing number of bursts per interval, and declined as the coherent components were increasingly perturbed in frequency. Further, for a fixed value of the ratio n/m discriminability was relatively independent of m. A model incorporating multichannel filtering and an optimum decision rule was reasonably successful in accounting for the experimental results.

Contribution of different frequency regions to detection of additive noise in vowels

An important issue in audio coding is the detection of quantization noise masked by speech. This investigation examined how different frequency regions contribute to the detection of broadband noise masked by a vowel. The vowel was synthesized by addition of the first 32 harmonics of a 200‐Hz fundamental, with amplitudes appropriate to /i/. Broadband noise was synthesized by random‐phase addition of the harmonics of a 10‐Hz fundamental that fall within the vowel bandwidth. Noise detection thresholds were measured in a 2IFC task with an adaptive procedure (Levitt, 1971). Detection thresholds were obtained for noise with spectrum envelope parallel to that of the vowel, and for the same noise containing −6.0‐dB spectral notches in either of the following bands: (a) 0.2–0.5 kHz; (b) 0.2–1.0 kHz; (c) 4.2–5.0 kHz; and (d) 4.7–5.0 kHz. Thresholds obtained in three highly trained listeners ranged from −27 to −32 dB (expressed in terms of noise‐to‐vowel power ratio). Threshold differences between spectral notch conditions were small. The results were compared to predictions of the excitation pattern model proposed by Moore and Glasberg [Hear. Res. 28, 209–225 (1987)]. [Work supported by OCAST‐HR4‐064.]

The Case for Using Nontraditional Judges in Forensic Contests

The forensics community should reintroduce the use of nontraditional judges in forensics tournaments. Diversifying the judging pool, while challenging tournament directors, will improve the educational value of forensics. The addition of nontraditional judges also will strengthen public support of the activity, and help forensics increase cultural diversity. This essay will build the case for using more nontraditional judges in forensic contests. I will discuss three benefits of increased use of nontraditional judges and consider some workability issues connected with their use. TRADITIONAL VERSUS NONTRADITIONAL JUDGES: HOW BEST TO EDUCATE STUDENTS? We may define a nontraditional forensic judge as a person with either limited training in contest judging or limited current experience in judging. Authors sometimes describe judges with limited experience as lay judges. Lay judges typically did not participate in forensics while students. People with limited current experience may include competitively inactive former competitors or forensics teachers. The common characteristic of both groups is their relative unfamiliarity with the nuances of current forensic practices. They may understand the structures of forensics contests but are not familiar with how situational norms affect actual practice. The issue of judging qualifications in forensic contests is neither recent nor trivial. Cox and Honse (1991) describe the long running discussion, particularly in the debate community, regarding the necessity of using trained listeners to judge debates. This discussion has centered on two questions: determining the goals of contest speaking; and identifying the necessary qualifications of a judge to achieve those goals. Most observers agree that contests include both educational and competitive aims. Students engage in contest speaking to improve their skills in areas like research, analysis and communication. Simultaneously, they are also competing against each other in pursuit of competitively driven outcomes, such as awards and other recognition. The role of competition is to provide an experiential learning framework for students to test and improve their learning skills. Competitive and educational goals in forensics often conflict. Competition sometimes causes people to short-cut appropriate educational behavior while pursuing competitive goals. This tension between competitive and educational goals underscores the important roles played by contest judges. Bartanen (in press) identifies three functions of the contest judge: educator, referee, and trustee of the activity. As an educator, the judge must help students learn research, analysis and communication skills by providing competent and appropriate comments about the content of their speeches or arguments. The educational function requires the judge to understand and apply the fundamentals of debate or individual events and to co-gently identify areas needing improvement. As a referee, the judge must appropriately apply contest rules and render a fair and impartial decision. This referee function requires the judge to understand both the structural (contest rules) and situational (nuances of competitive speaking and debating) elements of competition. As a trustee, the judge is responsible for maintaining the ethical and aesthetic standards of the activity. The judge must represent the larger community and apply aesthetic, ethical and rational standards that reinforce the utility of forensics as rational discourse. The judge, in other words, insures that debaters and speakers employ (in Wallace's use of the term) good reasons. Educators disagree about the necessary qualifications of people to fulfill these roles. Many educators believe that people trained in competitive forensics are the best contest judges. Robert Gass (1988), for example, advocates the use of trained judges who can promote high educational standards. Other critics (cited by Cox and Honse) oppose nontraditional judges because they believe, usually based on anecdotal evidence, that those judges evaluate contest winners randomly or only on contestants' delivery skills. …

Directional guidance method for the blind using time separation pitch discrimination of triple pulse signals

A guidance system for the blind, using time separation pitch (TSP) was simulated. Listeners estimated a simulated straight line course perpendicular to and bisecting a line between two horizontally separated transmitters, one transmitting a pulse doublet and the other, a single pulse. The course was represented by the coincidence of the single pulse with the doublet's center, forming a triplet of identical broadband, evenly spaced pulses. In test signals simulating deviations from the course, interpulse intervals were varied by changing the timing of the second pulse. The subjects' ability to discriminate pulse separations in triplets was measured. Next, subjects manipulated the timing of the second pulse of test signals to produce a triplet containing two equal intervals. Subjects' ability to discriminate and manipulate signals suggests that a trained listener could correct veering by observing changes in TSP.

Influence of adjacent pitch accents on each other's perceived prominence: two contradictory effects

In a previous study by Gussenhoven & Rietveld published in the Journal of Phonetics in 1988, in which listeners rated the prominence of the second pitch accent in synthetic utterances containing two pitch accents, it was found that decreasing the F0 on the first pitch accent peak (P1) induced lower prominence ratings on the second (P2). The present study replicates and extends this somewhat unexpected finding. Two groups of naïve subjects rating two slightly different sets of materials showed the Gussenhoven-Rietveld effect (i.e., they gave lower ratings of prominence for P2 as the F0 of P1 decreased). However, they did so only when P2 itself was below about 145Hz in a male speaking range. For higher values of P2 the Gussenhoven-Rietveld effect was reversed: lower prominence ratings were given to P2 as the F0 of P2 increased. Moreover, a group of trained listeners (phoneticians and speech researchers) failed to show the Gussenhoven-Rietveld effect at any level of P2. A broadly phonological explanation for the effect (and for its failure to occur under certain conditions) is proposed, such that, in normal range, the prominences of accents are not assessed individually, but are evaluated as a group to assess the utterance's overall degree of emphasis.

Luistervaardigheidstraining met Video en Wincalis

This paper discusses the possibilities of multimedia in language teaching and describes some example lessons for training listening comprehension with video. The lessons were created with WinCalis. The most important thing that was said about the teaching of listening comprehension, was that the training of listening should resemble real-life listening situations (Ur, 1984). When we listen we usually have a purpose and expectations. We listen actively to what we hear, make active use of context and use our own knowledge to understand what we hear. We usually listen to short chunks of language and are able to see the person speaking and the situation around him. If a response is needed, it is given directly after listening and this response is also reacted to immediately. Video can be used for training listening, because video allows listeners to see the person speaking and his surroundings, and because it gives an enormous amount of visual support. Two programmes were used for the lessons: a show about animals and a programme with magic and tricks. The lessons are meant to illustrate the maximum level of the Core Curriculum. Therefore the criteria mentioned in the core objectives are very important. The texts have to be authentic and use simple language. The texts used for the lessons are taken from -programmes for children and conform to these criteria. Still, the texts are relatively difficult, so the tasks that go with them should be kept fairly simple. The questions ask for information that is mentioned explicitly in the fragments or students are asked to think of their own solutions. It is shown how the pedagogical principles for training listening comprehension are incorporated in the lessons. Students get prepared as to the subject of the lessons so they know what to expect. They are told what to listen for before they see each fragment, so they have a purpose in listening. Feedback is provided immediately after each question is answered. Everything has been tried to make the lessons fun and to make the students listen actively. Trying to solve the tricks is so amusing and interesting that you cannot help but watch and listen carefully. Hopefully, students will realize that they can easily understand and enjoy an English programme on television.

Assessment of Speech in Children After Repair of Isolated Cleft Palate

The speech of 31 consecutive patients with isolated cleft palate was evaluated when they were between 10 and 14 years of age. All the children had Wardill-Kilner push-back repairs at a mean age of 7.9 months. Several speech variables were assessed by two trained listeners. In addition, an overall evaluation of the quality of the patients' speech was made by the listeners and by the patients themselves. The patients had some remaining speech problems, mainly hypernasality which was moderate or severe in 7 (23%). The prevalence of compensatory articulations was low, however (n = 4, 13%), and most patients thought that their own speech was normal or relatively normal.

Glottal fry and voice disguise: a case study in forensic phonetics

In recent legal proceedings, forensic phoneticians were called upon to analyse a tape-recorded message intended for the blackmail of a bank manager following the kidnap of his wife. The brief was to establish the likelihood that the tape recording may have been made by any one of three suspects, samples of whose speech were also made available. The comparison was greatly complicated by voice disguise employed by the speaker who recorded the kidnap tape. This disguise comprised a form of phonation described phonetically as ‘glottal fry’ or vocal ‘creak’. This form of phonation occurs normally in normal speech, but it has received most attention in relation to voice pathologies. On the other hand there are few references to its use as a form of voice disguise. This paper discusses the nature of the creak, and examines its effectiveness as voice disguise. In addition, a method is described for speaker identification regardless of the disguise. Results indicate that trained listeners without repeated presentations or instrumentation are able to match speakers with 65% accuracy when one voice is creaky, compared with 90% accuracy for undisguised voices. Using a Euclidean metric to compare the power spectra of the [s] sound, we find that creaky disguised voices may be correctly matched with the undisguised voice of the same speaker (9 distracters) in 5 cases out of 10. However, when the computer's task is made more similar to the perceptual task, selecting one speaker out of two, it achieves an accuracy of 81%. Implications for forensic phonetics are discussed.

Training native English speakers to identify Hindi dental and retroflex consonants

Training listeners to perceive consonantal contrasts that do not occur in their native language has proved to be difficult. Cross-language training studies usually produce about 10% improvement in performance. This improvement has not transferred to related material in different linguistic contexts. The present research had four aims: (1) to investigate the difficulty experienced by American-English listeners in identifying Hindi dental and retroflex stop consonants in different voicing conditions, (2) to test a new, computer-based, interactive training method, (3) to examine transfer of training to new voicing conditions, to a new vowel context, and to the voice of a new speaker, and (4) to test the hypothesis that increasing stimulus variability (in this case, training with one versus two voicing conditions) increases transfer of training. Subjects had differential difficulty identifying dental versus retroflex consonants produced in different voicing conditions. Further, this relative difficulty was affected by which speaker produced the contrasts and, to a lesser extent, by the vowel context. The computer-based training improved subjects’ consonant identification. However, this improvement showed little transfer to new stimuli. Finally, increasing stimulus variability during training did not affect transfer of training. [Work supported by NIDCD.]

Perceptual comparison of neoglottal, oesophageal and normal speech.

The sex, normality, intelligibility, rate, rhythm and intonation of 44 neoglottal, oesophageal and normal speakers have been judged by a panel of 10 trained listeners. It was found that the sex of the alaryngeal speakers was perceived correctly much less reliably than that of the normal speakers. Neoglottal speakers were rated more normal and intelligible than oesophageal speakers. The speaking rate of neoglottal speakers was judged not to be significantly different from that of normal speakers. Neoglottal speakers were considered more fluent and to have better intonation than the other alaryngeal speakers. Thus neoglottal speakers were found, on average, to be as good as or better than good oesophageal speakers in each of the respects of which judgements were made.

Discrimination of redundant spectral-temporal cues. II. Level and duration effects.

The effectiveness of stimulus level and duration on listeners’ ability to attend to and discriminate redundant increments in frequency and duration of pure tones is studied. The stimuli comprised pairs (T1, T2) of 1500-Hz tone bursts separated by a 60-ms silent interval. Sensation level was either 66 dB SL for both tones, or one of the tones was attenuated 25 or 40 dB relative to the other 66-dB SL tone. Tone duration was either 80 ms for both bursts, or T2 was two or three times longer than the 80-ms T1 burst. In the same three-interval, 2AFC task described in the above paper, listeners discriminated a ‘‘standard’’ pair of tone bursts from a ‘‘comparison’’ pair containing increments in the duration of T1 and/or in the frequency of T2. Data were obtained for eight highly trained listeners with normal hearing sensitivity. In three listeners, better discrimination performance was obtained for the increment occurring in the louder or longer burst. In four other listeners, tone level and duration had little or no effect on discrimination of either kind of increment. One listener failed to discriminate both increments in practically all conditions. Another source of interindividual variability is the listener’s ability to shift attention from one kind of increment to the other. [Work supported by OCAST Grant HSO-005 and Presbyterian Health Foundation.]

Discrimination of redundant spectral-temporal cues. I. Interindividual differences.

Interindividual differences in speech-processing ability of hearing-impaired listeners have been attributed to differences in the ability to process residual sets of highly redundant speech cues (e.g., Walden, 1984). This paper describes interindividual differences in listeners’ ability to discriminate redundant increments in the frequency and duration of pure tones. The stimuli comprised pairs (T1, T2) of 1500-Hz, 80-ms, 66-dB SL tones separated by a 60-ms silent interval [B. Espinoza-Varas, J. Acoust. Soc. Am. 74, 1687–1694 (1983)]. In a three-interval, 2AFC task, listeners discriminated a ‘‘standard’’ pair (interval 1) from a ‘‘comparison’’ pair (interval 2 or 3) containing increments in the duration (ΔT) of T1 and/or in the frequency (ΔF) of T2. Discrimination thresholds were interpolated from psychometric functions for ΔT alone, ΔF alone, and combinations of ΔT and ΔF. In highly trained listeners with normal hearing sensitivity (n=8), three main patterns of performance were observed: (a) multicue processing with accurate discrimination for both ΔT and ΔF (n=2); (b) single cue processing (selective attention) with accurate discrimination for only one kind of increment, usually ΔT (n=4); and (c) chance discrimination for both ΔT and ΔF (n=2). There are large interindividual differences in listeners’ ability to attend to and discriminate redundant increments in frequency and duration. [Work supported by OCAST Grant HSO-005 and Presbyterian Health Foundation.]

Modulated noise reference unit (MNRU) tests using DRT, DAM, and spelling alphabet test materials

The modulated noise reference unit (MNRU) has been widely used for reference conditions in evaluating voice coders for telephone applications. Opinion scores using the mean opinion score (MOS) are referenced to equivalent modulated signal‐to‐noise level. The DoD Digital Voice Processor Consortium has conducted extensive tests of voice processors using the Diagnostic Rhyme Test (DRT) to evaluate intelligibility and the Diagnostic Acceptability Measure (DAM) to evaluate acceptability. Spoken materials for the DRT, the DAM, and the ICAO spelling alphabet words and digits were degraded by speech‐modulated noise at levels from + 30 to − 15 dB. The resulting tapes were evaluated by trained listeners. Even small amounts of noise degraded acceptability, DRT intelligibility degraded more slowly, and the spelling alphabet was moderately intelligible even at the worst speech to noise levels. Relations among the different types of tests will be discussed.

Estimating the reliability of voice quality judgments: Comparing scales and statistics

Because perceptual measures are the primary means of assessing voice quality in the clinic, the reliability of listeners' judgments is of central importance. However, there is no consensus regarding the stability of trained listeners' judgments under conditions that mimic actual clinical practice, or as to the best method for estimating the reliability of such judgments. The present research compares the reliability of experts' estimates of vocal roughness gathered using equal appearing interval and visual analog scales, in order to determine (1) whether one scale type is inherently more reliable than the other; (2) the most appropriate methods for estimating the reliability of voice rating data; and (3) how reliable listeners actually are when they rate voices using these scales. The advantages and drawbacks of the many statistics used to evaluate reliability data will be discussed, and recommendations for standardizing perceptual measurement procedures will be presented.

Melodic character versus voice position in the perception of a two-voice musical structure

This study examines whether melodic character or voice position is a more influential factor in professional musicians' perception of a two-voice musical structure. A previous study [P. Baczewski and R. N. Killam, J. Acoust. Soc. Am. Suppl. 1 86, S17 (1989)] examined 15 highly trained listeners' response to a stimulus consisting of a 16-measure piece of music. They were presented with a taped performance (prepared by professional musicians) of three repetitions of a Mozart duet, together with the experiment's instructions. Experimental design allowed measurement of response to each stimulus repetition. A difference was noted in the perceptual strategies for notation of top and bottom lines of the duet. This current study examines a repetition of the above experiment in which the positions of the lines are reversed (transposition of the bottom line into the range of the top and vice versa). This procedure yields insight into whether melodic character was responsible for the lesser degree of total correct response for the lower line of the previous experiment, or whether lower lines are inherently less well perceived and notated by musically trained subjects. [Work supported in part by an Apple Education Foundation Grant.]

Short-Term Stability Measures for the Evaluation of Vocal Quality

The vocal quality of 64 normal subjects and 57 subjects suffering various degrees of glottal cancer was investigated using acoustic measures of six different aspects of the voice signal: tone period perturbation, amplitude perturbation, waveform perturbation, vocal noise, spectral periodicity and spectral distortion. The measures were estimated taking the glottal cycle as temporal reference unit to make the influence of the differences in tone period from one person to another as low as possible. The measures were evaluated with regard to (a) their ability to discriminate between healthy and sick subjects, and (b) their correlation with the perceptual evaluation of four trained listeners. The results suggest that signal processing techniques are unsatisfactory for clinical diagnoses but useful for monitoring voice quality.

Perception of a musical construct by professional musicians

This study examined 15 highly trained listeners' response to a stimulus consisting of a 16‐measure piece of music. Subjects were music performance graduate teaching fellows. They were presented with a taped performance (prepared by professional musicians) of three repetitions of a Mozart duet, together with the experiment's instructions. Experimental design allowed measurement of response to each stimulus repetition. Subject responses were recorded in one of two differing formats: eight subjects recorded the musical notation of each of the three stimulus repetitions in a different color of pen; seven subjects recorded the musical notation of each of the three stimulus repetitions on a different translucent paper overlay. Subjects were allowed 3 min to record each response. Data were encoded and analyzed to provide measurement of cumulative response arising from all three repetitions. Effects of pitch boundary and constituent groupings were obtained. A model emerges of complex perceptual strategies employed by musically sophisticated subjects. The strategies incorporate pitch relationships, contour, and grouping. [Work supported, in part, by an Apple Education Foundation Grant.]