MSc Speech and Hearing Sciences  - Dissertation Abstracts

This page contains abstracts of some of the student project dissertations undertaken as part of the MSc in Speech and Hearing Sciences.


Beat Detection and Specific Language Impairment: How is Sensitivity to Amplitude Envelope Onsets related to Speech, Language and Literacy?

Chloë Bate (2003)

This study looked at “beat detection” (sensitivity to changes in amplitude envelope onsets) in a group of 12 children with language impairment (LI) and 11 control children matched on age and non-verbal IQ. This auditory processing ability is argued to be a non-speech-specific mechanism for separating onsets and rimes and therefore, an important factor in the development of phonological awareness. A deficit in this ability has been postulated as the primary deficit in development dyslexia and it was argued that a serious deficit could lead to spoken as well as written impairments of the kind ground in specific language impairment. Thus, the speech, language and literacy abilities of the children participating were assessed to investigate a possible relationship between these and beat detection ability. Results showed that a significant proportion, though not all, of the LI children exhibited the deficit and these were the ones that generally had the most problems with literacy and phonological awareness. Although the beat detection deficit did not seem to be a sufficient cause of the speech and language problems seen in theses children, there did seem to be a close relationship between beat detection ability and phonological awareness and literary skills. This suggests that a beat detection deficit may be a contributing or aggravating factor to the phonological awareness and, thereby, literacy problems seen in many of the children with LI in this study.

Can Normal Listeners Adapt to an Upward Asymmetric Spectral Shift? Implications for Bilateral Cochlear Implants

Lucinda Saul (2003)

Bilateral cochlear implants may lead to binaural advantages that enhance speech perception and localisation difficulties. However, in bilateral cochlear implant users electrode array insertion lengths are likely to differ between the two implants, such that they must overcome limitations imposed by the difference in frequency-to-place mismatch to obtain these advantages. The spectral shift experienced in each ear will differ, generating an asymmetric shift between the two ears. Adaptation to these spectral shifts has been demonstrated in simulation studies, but adaptation to a dichotic spectral shift across the two ears has yet to be investigated. In this study, the effects of an upward asymmetric shift, with the equivalent of a 6mm basalward shift on the basilar membrane in one ear, has been stimulated and performance tested under binaural and monaural conditions. A sine wave vocoder simulated bilateral cochlear implants with a total of six frequency channels. There was a dichotic presentation of the frequency information to the two ears from the six input analysis bands, so that the three odd frequency sine waves generated were assigned to one ear and the even frequency sine waves were assigned to the contra-lateral ear. Thus, subjects would benefit from integrating information across the two ears and would not attend to one ear alone. After nine 40-minute sessions of connected discourse tracking with the dichotic shifted condition, speech perception performance improved significantly in the trained dichotic shifted, monotic odd-channels shifted and unshifted conditions. However, listeners were unable to optimise their speech perception by binaural integration. In post-training sentence perception tests, there was no significant difference between the dichotic shifted condition and the monotic even-channels unshifted condition, suggesting that listeners were only attending to their better ear. Although normal listeners were unable to effectively integrate the binaural information, it is clear that they were able to adapt to the trained condition to some extent, but it is unclear exactly what they were learning from it. It is possible that listeners require longer exposure to the asymmetric spectral shift to adapt and to achieve binaural integration.

The Resolution of Lexical Ambiguity arising from Place Assimilation: Evidence from a Gated Speech Task

Sarah Bateman (2002)

Place assimilation in English can give rise to lexical ambiguity, as in the sentence “There’s your ca[p] by the door”, where ca[p] could be interpreted either as ‘cap’ or an assimilated form of ‘cat’. The experiment reported here used a gating task to investigate the timing and effectiveness of different cues used by listeners to resolve this type of ambiguity. It was found that in a neutral sentence context, listeners were more likely to identify the ambiguous word as a word directly matching the surface form (non-coronal), although some identified it as a word in assimilated form (coronal). A constraining interpretive context favouring the coronal word increase the proportion of coronal word candidates, although it did not reduce the time needed to identify ambiguous words with confidence. When acoustic information about an ambiguous word was most salient – at word offset and  before the subsequent word was fully identified – there was a strong perceptual preference for the word most directly matching the surface form. Contrary to a prediction that a phonological context legitimising assimilation should allow for more coronal candidates, such a context appeared instead to strengthen the acoustic information about the non-coronal place of articulation, thereby increasing the proportion of non-coronal candidates. This effect was shown even when a preceding interpretive context made the non-coronal candidate highly implausible. The number of coronal candidates increased only when further lexical information became available. These findings are discussed in the light of predictions derived from psycholinguistic and engineering models of spoken word recognition.

The Importance of Pitch Information in Speech Recognition with reference to Mandarin Chinese

Yun-Chin Chou (2002)

The aim of this study is to investigate the contribution of pitch information to the performance of Mandarin speech recognisers. In this project, a telephone speech database was used to build the recogniser. Different forms of pitch-based features: Fx, delta Fx, delta delta Fx, normalised Fx, and their different combinations were incorporated into the feature vectors in a series of experiments observing their effects on the Mandarin recogniser. The results show that delta Fx has the largest enhancement to the recognition rate, compared to other pitch-based features, while Fx as sole pitch feature causes performance degradation. Normalising Fx and clustering the data files of the same gender are helpful. The experiment result from trigram suggests that there is no obvious interaction between the language model and Fx features.

The Effect of Timbre on Pitch Perception in Pianists, Violinists and Non-Musicians

Rebecca Dumican (2002)

The auditory abilities of musicians compared to those of non-musicians have long been of great interest and have sparked a large amount of research in hearing science, psychology and other related fields. The focus of this research has often been on the ability of musicians to outperform non-musicians in frequency discrimination tasks and the effect of timbre on pitch perception in all populations. How musicians have become better at frequency discrimination tasks is an important issue and researchers have suggested various explanations. Musicians may have an improved overall auditory capability, or their skill may lie in being able to ignore the interference of ambiguous characteristics of a tone such as timbre. Timbre is reported to have a deleterious effect on pitch perception in both musicians and non-musicians. There is some evidence to suggest that “bright” or “dull” timbres can produce a bias in the pitch percept. Some research into the skill of musicians has found that there may be an effect of musical discipline on the ability to perform fine frequency discrimination tasks. For example, the experience of tuning a musical instrument regularly may enhance this ability. To this end, three subject groups: non-musicians; violinists and pianists were compared for the thresholds they obtained in a set of frequency discrimination tasks. The effect of timbre upon these frequency discrimination thresholds was assessed using some tones of markedly different timbre. It was found that the pianists and the violinists produced almost identical thresholds and that both groups of musicians outperformed the non-musicians in the set of frequency discrimination tasks. All subject groups found timbre to be a factor which interfered with frequency discrimination. However, the musicians found timbre differences less distracting than non-musicians. This shows that musicians cannot ignore timbre, but that they may utilise the similarity between two tones in order to produce an accurate pitch estimate. Tones with a particularly “bright” or “dull” timbre did produce a bias in the pitch percept, subjects hearing them as higher or lower in pitch respectively. In this experiment, experience of tuning [a musical instrument] had little effect on frequency discrimination thresholds, suggesting that musical experience alone accounts for the enhanced ability, not the specific training of a musical discipline.

Single-Word Processing by Japanese-English Bilinguals: A 3T fMRI Study

Ryoko Matsumoto (2002)

Neural representation of two different types of orthography: Japanese kanji and the English alphabet was investigated using fMRI. The subjects were Japanese-English late bilinguals and were shown two-character kanji words and their translation in English in single word and text conditions. As previous kanji reading studies have shown contrasting activation data, probably because of inconsistent criteria in the choice of stimuli, we used only concrete nouns for both Japanese and English stimuli.  The results revealed that reading Japanese words activated the left interior insula and the right precuneus, indicating that the subjects used subvocal articulation and recognised the words more as images. When reading English words, the left fusiform gyrus and bilateral cingulate gyri were activated, which suggests that the subjects tried to decode the sequence of letters and load them from memory. Presenting multiple words in text conditions revealed that horizontally-presented texts induced smoother eye movements. Significant activation was also observed in the cerebellum, indicating that it serves for non-motor linguistic functions as well.

The Plasticity of the Brain regarding Auditory Perception – A Project Using Spectrally-Rotated Speech

Ruth Finn (2001)

There is much evidence to suggest that listeners can learn to perceive speech that is so distorted as to be unintelligible initially. During the course of this study, four subjects underwent a training programme following the Continuous Discourse Tracking (CDT) method, using spectrally-rotated speech (an extreme distortion, initially studied by Blesser (1969)). Each subject completed six hours of this training (eight 45-minute sessions) and a series of speech perception tests. Results from these subjects were compared with results from four controls who only carried out the perception tests. It was shown that only subjects trained to perceive spectrally-rotated speech had an increased ability to do so. In sentence tests with a female speaker and only audio information, performance by the trained group increased 5% mean words correct (after no training) to 27 mean words correct (after 3 hours of CDT) and finally to 42 mean words correct (after 6 hours of CDT). Similar results were found from tests using a male speaker and audio information alone: 9 mean words correct initially, increasing to 18% after 3 hours of CDT, and 26% correct after 6 hours of CDT. The performance of the trained group also showed significant improvement in audio-visual sentence tests: 34% mean words correct initially; 74% correct after 3 hours of CDT and 83% correct after 6 hours of CDT. Although it is unknown whether a complete adaptation to the spectrally-rotated speech is possible, it is clear that significant progress can be made over a relatively short period of time.

The Evolutionary Nature of Emotion: Vocal Cues to Happiness

Victoria Maschio (2001)

The aim of this study is to investigate the nature of five positive emotions which are distinguishable by voice and not facial expression. These are Amusement, Achievement, Relief, Contentment and Sensual Pleasure. The two theories tested are the evolutionary perspective of “basic” emotions proposed by Ekman and Russell’s theory of emotional space. The stimuli used were non-verbal expressions of emotion of short duration. The study is divided into three parts: a categorisation test investigating Ekman’s hypothesis; a multidimensional scaling test to try and create a similar model to Russell’s and lastly an observational acoustic analysis of the five better-recognised stimuli. The results showed that recognition of the emotions was well above chance level and that Amusement, Achievement, Relief and a form of Pleasure are indeed basic emotions. It was also revealed that these emotions are not related to Russell’s model of emotional space. The acoustic analysis showed that features such as voice quality and fundamental frequency trace can differentiate these emotions, even though the evidence obtained was not conclusive.

An Acoustic, Larynographic and Perceptual Study of Opera and Belt Singing Voice Qualities

Esther Shanson (2001)

This study examined the perceptual, acoustic and laryngographic differences between opera and belt voice qualities, as defined by the physiological specifications of the Estill model. The voice qualities were recorded on three pitches and four vowels by Estill-trained sopranos. The perceptual tests on these recordings were played to twenty naïve and twenty expert subjects.  The perceptual difference between opera and belt was found to be significant. The voice qualities were more reliably distinguished by the expert subjects. Differences between the voice qualities were less obvious at higher pitches and on certain vowels. Front and open vowels on A4 were most suited to belt quality. The closed quotient values were considerably higher for belt, rather than opera quality. Higher spectral harmonics were enhanced in belt, whereas the fundamental frequency had a particularly high intensity in opera quality. In both qualities, there was evidence for the singer’s formant. Examining the same data on acoustic, larynographic and perceptual levels gave a more complete insight into these voice qualities.

Text-to-Speech (TTS) Synthesis Evaluation: A Study on the Reliability of the ITU-T Standard Recommendation P.85 for the Evaluation of English Synthesisers on the Web

Yolanda Vasquez-Alvarez (2001)

An evaluation was conducted on six TTS systems for English in order to study the reliability of the ITU-T Standard recommendation P.85. This recommendation is based on subjective judgements on a number of rating scales for individual TTS systems. This recommendation was compared with a much simpler method, i.e. a Pair Comparison (PC) test, based on pair comparison preferences. Four different genre materials were used in both tests, i.e. E-mail, news, catalogue entries and name-address-phone numbers which were considered representative of the kind of texts which technological applications will have to deal with. The final aim of this study was to compare the validity of both methods. It was found that the results of both tests were consistent and few differences were observed. Therefore, it was concluded that the ITU-T appears to be as reliable a method as the PC test in terms of retest reliability. However, it presents many problems with the scales and the interpretation of results when using Mean Opinion Scores (MOS). It is also less easy to run because participants are exposed to much more information which they need to retain when making their judgements. It must, however, be remembered that there  are some constraints to a project run on such a small scale. Therefore, future work should be carried out in order to  confirm, extend or reject the findings presented in this study.

A Computer-Based Auditory Training of English Contrasts for Cantonese Speakers

Elizabeth W.W. Chua (2000)

Many studies have shown that perceptual training with appropriately selected prototype stimuli would enhance L2 learners’ ability to identify new non-native speech sounds. Using the CSLU Toolkit, this project attempts to design a computer-based auditory training module that helps Cantonese speakers to identify some difficult English voicing contrasts more accurately. Our training procedure has followed Jamieson’s guidelines (1995) and employed the perceptual fading technique to redirect listeners’ attention to the critical duration cues of fricative voicing contrasts. This method has proved to be efficient in improving listeners’ identification accuracy. After a one-week period of training, the subjects have shown significant improvements in perception, from pre-test to post-test.  However, training did not seem to transfer from the trained to the untrained positions. Moreover, only a small generalisation effect was found in our results. The limited success in the transfer of training is mainly due to the inadequate variety of training stimuli and number of speakers in our module. These findings reassert the important roles of stimulus variability and talker variability in designing an effective auditory training programme. In terms of the subjects’ evaluation of this training module, they regard it as being very useful in helping them to learn the different contrasts. Most feel more confident in identifying the non-native speech sounds after completing the training. On the whole, this study has demonstrated the viability of using an auditory training in L2 pedagogy.

Speech Synthesis by Word Concatenation: An Investigation into the Relative Importance of Intonation, Duration and Speech Segment Recording Conditions on the Preferences of Listeners

Mary Jane Paly Vera (2000)

This work is concerned with speech synthesis by the concatenation of speech segments and more specifically, about how the process of unit selection in concatenative synthesis systems could be improved. Unit selection is very important for the production of better quality synthetic speech. By choosing better speech segments from a speech database, concatenative speech synthesis systems can minimise signal processing such as smoothing and prosody manipulation of concatenate speech segments and, therefore, they can introduce less distortion in the final synthesis.

In this study, the relative importance of intonation, duration and speech segment recording conditions on listeners’ preferences was investigated. However, these were just some of the factors considered during the unit selection process. An experiment was designed in which the relative importance of these factors was tested against the perception of telephone users. Telephone users were the subjects of this study. since synthetic speech produced by concatenative systems is mainly used in over-the-telephone applications. For this reason, this area demanded further investigation. In addition, we chose to work with words as the speech segment for our speech synthesis. After creating several synthetic utterances, a perceptual test was run. The results obtained were used to create a rank ordering for the three different factors that we were investigating. The results of the experiment showed that intonation was the factor that had the greatest influence on the listeners.

A Spreading Activation Model of Word Selection

Maria Ovens (1999)

A three-layer spreading activation model of word selection was developed within the framework established by Dell’s 1988 connectionist model of speech production. The model uses either semantic or phonological information, or a combination of the two, to select a word for production. A proposed mechanism for word priming, residual activation in the phonological and semantic layers, was investigated. Experiments were carried out which demonstrate that the model is capable of reproducing facilitation and inhibition effects found in human subjects, but only at the expense of setting a large number of arbitrary parameters. In addition, further work is suggested, which might go some way towards alleviating this problem, and which might allow the model to be used for stimulating other psycholinguistic phenomena.

Phonological Word and Stuttering: A Contrastive Study between English and Spanish

Isabel Vallejo Gomez (1999)

Stuttering, a disorder of oral language expression, has been studied extensively for decades. Throughout this time, different explanations have been offered to account for this kind of speech disorder, from theories that root the cause of stuttering in the motor-dynamic aspects of speech to others which suggest a psychological orientation to this condition. This study deals with stuttering and the concept of phonological word in English and Spanish. Two lines of research, one in psycholinguistics and the other in linguistics are combined in a comparative study which attempts to offer a plausible explanation for the causes of stuttering and its underlying structure. The former will deal with the mental mechanisms involved in patterns of stuttering, while the latter will focus primarily on prosodic phonology, offering a theoretical framework in which to analyse the experimental data. The main purpose of this study, a continuation of the research proposed by Howell, is to discover whether there are universal stuttering patterns associated with linguistic structures or whether the pattern of stuttering is language-specific. Studies focussing on the pattern of stuttering in English have shown that the stuttering rate is affected by the function/content word distinction and their position within the phonological word. In this study, it is argued that since Spanish and English are not directly related languages, the stuttering patterns observed in English may not be applicable in Spanish. Although stuttering is a universal phenomenon, some of the linguistic characteristics it presents are essentially language-specific.

Speech and Non-speech Auditory Processing in Children with Dyslexia

Evdoxia Manganari (1998)

Experimental studies have shown that poor readers are inferior to good readers in some aspects of speech perception. Two hypotheses have been proposed to account for this deficit: i) a speech-specific failure in phonological development and ii) a general deficit in processing auditory information. The aim of the present study was a) to investigate the perception of plosive consonants in dyslexic children using identification and discrimination tests, and its relation to their phonological ability, b) to examine the specificity of a possible deficit by means of non-speech control stimuli consisting of the F2-transitions alone and c) to investigate the relationship between performance in speech perception and a masking task in an attempt to test the hypothesis put forward for abnormal auditory masking as the underlying cause of the difficulties experienced by language- and reading-impaired children (Wright, 1997). Two groups of children matched for age and non-verbal intelligence were selected to be well separated in reading and spelling ability. The two groups were found to differ significantly on the speech identification and discrimination tests, but not on the non-speech conditions. In addition, significant group differences were obtained in the backward masking condition, but not in forward and simultaneous masking. The investigation of the relationship between performances on the masking task and speech perception failed, however, to provide evidence of a possible link. Although the findings do not support the General-Auditory-Deficit Hypothesis as the underlying cause of the reading-impaired childrens’ difficulties, they clearly point to a deficit in auditory processing in dyslexic subjects.

Dysfluencies Exhibited by Stutterers: Life-span Differences in Loci and Dysfluency Type

Lea Pilgrim (1998)

Previous work has indicated that dysfluency types exhibited by stutterers do not remain constant across the life span (Conture, 1990). Other studies have found that the loci of dysfluencies also varies as a function of age (Brown, 1945; Bloodstein, 1974). Until now, the interaction of these two variables, word type and dysfluency type, has not been investigated within a single study. Here, the interaction of word type and dysfluency type is explored, within an examination of the differences these variables exhibit with age. In the current study, it has been found that the dysfluency type ‘part-word repetition’ is much more likely to be exhibited by younger stutterers and also that this finding rests predominately on function words (for example, prepositions, pronouns, modal verbs) rather than on content words such as nouns, verbs and adjectives. An interaction between word type and dysfluency type was also found; whole word repetitions are more likely to occur on content words. Finally, the proportion of prolongations exhibited on content words was compared to the proportion of part-word repetitions on function words. The results show a significant difference which rests on the youngest group of stutterers. These results identify several significant interactions between word type, dysfluency type and age. The findings are discussed within the framework of a new model of stuttering (Au-Yeung, Howell & Pilgrim, in press) and suggestions for future studies are also proposed in the discussion section of this paper.

The Identification of Reading and Spelling Development of Greek-speaking Children and Predictions about the Nature of their Reading Difficulties

Vassiliki Xydi (1998)

This study attempted to examine reading and spelling development as well as the development of cognitive measures of Greek-speaking children. These factors were examined before reading had begun (when the children had finished kindergarten); after initial instruction had taken place (once the children had completed the first grade of elementary school) and after reading instruction had been completed (the children had completed Grade 2). Results from the reading task revealed that Greek children display rather high reading accuracy and that their spelling was mainly phonetically acceptable; revealing the complexities of the Greek orthographic system. The results of the Phoneme Deletion task provided useful insights into the significance of phonological awareness in relation to reading and spelling. A picture-naming test showed similar results and Phoneme Deletion gave insights concerning output phonology and phonological working memory. Interestingly, Auditory Discrimination, a test often labelled “easy” and a predictor of reading achievement in other studies, showed very low correlations with all of the other measures. Analyses of the data clarified issues concerning the nature of reading and spelling development of children of a consistent orthography. Firstly, the findings demonstrated the inadequacies of a model such as Frith’s to account for reading and spelling in a regular orthography. Secondly, the findings encouraged the prediction that Greek dyslexic children may suffer from a phonological deficit, thus their type of dyslexia is characterised as “phonological dyslexia”. Lastly, the most important factors that predict reading and spelling achievement were found to be phonological processes, mainly those relating to output phonology and phonological working memory.

Assessing the Effects of Speech-reading on Speech Perception with Sensorineural Hearing Loss – Simulations in Natural Conditions

Peter Lege (1997)

This project might be regarded as a first attempt to produce video recordings that tried to stimulate sensorineural hearing loss. The stimulated hearing loss aimed to mimic presbyacusis with a mild-to-moderate sloping hearing loss in the high frequencies. Within this framework, the effects of speech-reading on speech perception with sensorineural hearing loss were investigated. Video recordings of a female speaker of Southern British English were made in natural rather than laboratory conditions, namely in a quiet office and in a park surrounded by streets with heavy traffic. In addition, binaural auditory recordings also helped to give a more realistic and natural interpretation of the recorded scene. The videotape produced may also give interested parties the opportunity to appreciate the impact of sensorineural hearing loss on speech. The two experiments performed employed five native English speakers with normal hearing. The first experiment used sets of VCV syllables and investigated the role of speech-reading with distorted and undistorted speech; with quiet and noisy backgrounds. Speech-reading did not raise speech intelligibility significantly when the speech signal was distorted. Surprisingly, quiet and noisy backgrounds did not make any significant differences in terms of correct consonant identification. One of the main reasons might be that, owing to the elevated hearing threshold, differences between quiet and noisy backgrounds were just reduced. It was also suggested that in previous similar investigations that real hearing-impaired people achieve lower speech recognition scores than unimpaired subjects listening to stimulated speech. A second experiment was designed rather like a pilot experiment and used two short stories with a distorted speech signal in an audiovisual and auditory-alone conditions. The results did not allow any reliable analysis. It turned out that the short stories seemed to be too different to allow a direct comparison of the results. It might be more appropriate to run pilot experiments in order to evaluate and obtain short stories which are similar in terms of content and difficulty. However, a tendency could be observed that speech-reading might be of benefit for the short stories. It was concluded that the stimulation of hearing loss needs to be evaluated by comparing it with hearing-impaired people. Also, additional stimulated hearing-aid amplification may help to clarify whether quiet-noisy background conditions are dependent on the actual hearing threshold level and to what extent spectral smearing and loudness recruitment play a role. If the videotape turns out to be at least a near-realistic simulation, it could then be used for presentation purposes with appropriate speech or text material.

Laryngographic Assessment of Voice Changes with Altered Hydration Status

Julia Selby (1997)

The purpose of this study was to investigate the effect of mild dehydration on voice quality. Eight healthy male subjects completed a screening questionnaire and underwent a brief ENT examination to check that there were no indications of laryngeal pathology or vocal abuse. They then fasted for a period of eighteen hours and were administered a BIA test to establish that their body fluids had decreased and that they were dehydrated. Voice recordings using the Electrolaryngograph were then made. These consisted of humming, producing sustained vowels at low, comfortable and high fundamental frequencies, reading a two minute passage and talking informally for two minutes. The subjects then drank two litres of an electrolytic drink and waited one hour to become rehydrated again. A second body fluid measurement was taken to determine that the subjects were rehydrated and the same test materials were recorded again. The vocal cords were viewed directly using the Laryngostrobe system when the subjects were dehydrated and rehydrated. Measures of vocal cord vibration irregularity, voice fundamental frequency range, fundamental frequency mode and vocal fold vibration jitter were made using the Laryngograph (Fourcin and Abberton, 1971). Hydration was found to increase the mode in the reading task (p=0.002). For all other measures, no significant effect of hydration was found. Theoretical discussion focuses on the possible explanations for this and on recommendations for continued research in this area.

Subjective and Objective Assessment of the Speech of Post-lingually Deafened Adults. What Makes People "Sound Deaf"

Caroline Newton (1995)

A combination of subjective judgements from trained phoneticians and objective measurements indicated deterioration in the speech of six subjects with an acquired post-lingual profound hearing loss, compared to the speech of six normally-hearing control speakers. Deterioration was found in areas including voice quality, rate, Fx range and fricative production, though a large amount of variation was observed between the different speakers. Results taken from the inexperienced listeners showed that they were able to distinguish the hearing-impaired speakers from the normally-hearing speakers. This was shown to be statistically significant in a two-tailed t-test for independent samples at the 0.05 level (p = 0.17 * 10-5). Pearson’s product-moment correlation coefficients were calculated from the results of the experienced listeners and, where appropriate, from those of the objective measures. From these results it was clear that the greatest correlation between these and the judgementthat someone “sounds deaf” particularly concerned alveolar frequencies (correlation 0.96). It was concluded that, as far as these samples of speakers and listeners are concerned, it is the deviant production of alveolar fricatives that makes someone “sound deaf”.