Number Of Talkers Research Articles

Development of an Arabic speech intelligibility test for children

EnAbstract Objectives Speech intelligibility is usually expressed as a percentage of spoken words, sentences, or phonemes correctly identified by a listener or a group of listeners when spoken by a talker or a number of talkers. Communicative disorders could have a deleterious effect on speech intelligibility. To date, there are no Arabic tests measuring speech intelligibility. Aim It is necessary to develop and standardize an objective Arabic speech intelligibility test that can be used to estimate the degree of deterioration of speech intelligibility in communicative disorders as an attempt to develop a valid and a reliable tool to assess the efficacy of different therapy programs for different communicative disorders. Patients and methods This study was carried out on 200 children with an age range of 4–12 years whose language skills were either fully developed or equivalent for age. All children were selected randomly from the outpatient clinic of phoniatrics and had any of five selected speech disorders affecting speech intelligibility. Each child included in the study was subjected to two evaluations: a subjective rating of the child’s speech intelligibility and the developed Arabic speech intelligibility test, which is meant to be an objective measure. Results The results showed a highly significant correlation between the scores of the Arabic speech intelligibility test and the average scores given by raters, the average scores for words, and the average scores for sentences (0.92, 0.98, and 0.84, respectively). Conclusion The developed test is proven to be valid and reliable for measuring speech intelligibility in children and could be categorically classified according to ranges of severity.

An Energetic Limit on Spatial Release from Masking

This study tested the hypothesis that energetic masking limits the benefits obtained from spatial separation in multiple-talker listening situations, particularly for listeners with sensorineural hearing loss. A speech target was presented simultaneously with two or four speech maskers. The target was always presented diotically, and the maskers were either presented diotically or dichotically. In dichotic configurations, the maskers were symmetrically placed by introducing interaural time differences (ITDs) or infinitely large interaural level differences (ILDs; monaural presentation). Target-to-masker ratios for 50 % correct were estimated. Thresholds in all separated conditions were poorer in listeners with hearing loss than listeners with normal hearing. Moreover, for a given listener, thresholds were similar for conditions with the same number of talkers per ear (e.g., ILD with four talkers equivalent to ITD with two talkers) and hence the same energetic masking. The results are consistent with the idea that increased energetic masking, rather than a specific spatial deficit, may limit performance for hearing-impaired listeners in spatialized speech mixtures.

Prediction of the effectiveness of a sound-masking system in an open-plan office including the Lombard Effect

Sound masking can improve speech privacy in rooms by increasing background-noise levels that mask distracting speech sounds. The Lombard Effect indicates that an increase in background-noise level can increase talker voice levels, reducing speech privacy and the benefit of a sound-masking system. To investigate this, a model of an existing open-plan office was created in CATT-Acoustics and validated. The model was used to predict speech-transmission index (STI) and the effectiveness of a sound-masking system, without and with the Lombard Effect, described by the existing Lombard Voice Model. Predictions were made for ambient-noise levels of 30, 40, and 45 dBA, at various distances from a primary talker, and for 0–4 secondary talkers. With 30-dBA ambient noise, STIs at 1 and 4 m varied with the number of talkers from 0.67 to 0.91 and 0.23 to 0.62 without the sound-masking system, from 0.58 to 0.70 and 0.13 to 0.25 with it but ignoring the Lombard Effect, and from 0.64 to 0.73 and 0.18 to 0.27 with the Lombard Effect. The Lombard Effect reduced the benefit of the sound-masking system by 3–9% and 7–22%. With higher ambient noise, the system is less effective; the Lombard Effect can almost cancel its benefit, resulting in increased STI (decreased privacy) with the system operating.

Listener judgments of age in a single-talker 48-year longitudinal sample

Numerous studies have demonstrated that listeners can make relatively accurate judgments of a talker’s age from hearing the talker’s voice. Materials in these previous studies have included sustained vowels (phonated and sometimes whispered), sentences, and passages of discourse. The number of talkers has ranged from 4 to 150, but in nearly all cases talkers were recorded only once. In the present study, the materials were recorded from a single talker who gave regular public speeches over a period of 48 years. Young adult listeners performed age judgments on samples extracted from 20 speeches chosen at 2–3 year intervals spanning the 48-year period, three samples per speech. Samples lasted 5 to 10 s and were chosen to minimize content that would identify the talker or link samples from the same speech to each other. In separate experiments, listeners listened to these 60 samples and after each one judged the talker’s age either by choosing from three categories (50–66, 67–83, or 94–100 years) or by making a direct age estimate. Accuracy of these estimates will be compared to previous studies and examined as a function of both the talker’s chronological age and acoustic measures performed for this talker in a separate experiment.

The influence of multiple narrators on adults’ listening comprehension

Research has demonstrated that variable, talker information—such as the number of talkers—affects listeners’ perception and processing of linguistic information during various laboratory tasks. In particular, the detrimental effects of multiple talkers are highlighted during online speech perception tasks with little contextual support [isolated word recognition; e.g., Mullennix et al. (1989), Ryalls and Pisoni (1997), Sommers and Barcroft (2011)]. Nonetheless, it is unclear how multiple talkers might affect listeners’ perception of linguistic information in more complex spoken language tasks utilizing real-time, fluent speech. The present experiments were conducted determine if information contributed by multiple talkers influences adults’ auditory story comprehension in the presence of both quiet and background noise. The accuracy and reaction time data did not support the hypothesis that talker information affects the perception of linguistic information during auditory story comprehension. Thus these data bring to light theoretical perspectives that emphasize the importance of looking across experimental tasks to better understand talker-specific information’s pattern of influence on spoken language processing [e.g., Sommers and Barcroft (2006), Werker and Curtin (2005)].

Listening to speech in a background of other talkers: Effects of talker number and noise vocoding

Some of the most common interfering background sounds a listener experiences are the sounds of other talkers. In Experiment 1, recognition for natural Institute of Electrical and Electronics Engineers (IEEE) sentences was measured in normal-hearing adults at two fixed signal-to-noise ratios (SNRs) in 16 backgrounds with the same long-term spectrum: unprocessed speech babble (1, 2, 4, 8, and 16 talkers), noise-vocoded versions of the babbles (12 channels), noise modulated with the wide-band envelope of the speech babbles, and unmodulated noise. All talkers were adult males. For a given number of talkers, natural speech was always the most effective masker. The greatest changes in performance occurred as the number of talkers in the maskers increased from 1 to 2 or 4, with small changes thereafter. In Experiment 2, the same targets and maskers (1, 2, and 16 talkers) were used to measure speech reception thresholds (SRTs) adaptively. Periodicity in the target was also manipulated by noise-vocoding, which led to considerably higher SRTs. The greatest masking effect always occurred for the masker type most similar to the target, while the effects of the number of talkers were generally small. Implications are drawn with reference to glimpsing, informational vs energetic masking, overall SNR, and aspects of periodicity.

Sentence recognition as a function of the number of talkers in competing multi-talker babble

Multiple-talker babble is used in studies of speech perception and processing both as a tool for loading perceptual and cognitive tasks and for direct assessment of speech perception in noise. Single interfering talkers and babbles containing few talkers differ dramatically, however, from babbles with high numbers of talkers. While low N-talker babbles provide greater opportunities for glimpsing speech targets during spectral and temporal dips in the maskers (i.e., less energetic masking), the linguistic information available to listeners in the maskers themselves also imposes higher-level interference (i.e., informational masking) that may detract from the identification of target speech. In order to assess the relative overall masking effects of N-talker babble for sentence recognition, the current study utilizes a range of maskers varying in N (1-8), as well as speech-shaped noise. Data collected to date show that performance declines as talkers are added to the masker (that is, as energetic masking increases), but that performance in 6-8 talker babble is significantly better than in speech-shaped noise. Individual variability on speech perception in these various maskers is being assessed by comparing performance on the speech-in-noise task to performance on a range of cognitive and psychoacoustic tasks.

Auditory externalization in hearing-impaired listeners: The effect of pinna cues and number of talkers

Hearing-aid wearers have reported sound source locations as being perceptually internalized (i.e., inside their head). The contribution of hearing-aid design to internalization has, however, received little attention. This experiment compared the sensitivity of hearing-impaired (HI) and normal-hearing listeners to externalization cues when listening with their own ears and simulated behind-the-ear hearing-aids in increasingly complex listening situations and reduced pinna cues. Participants rated the degree of externalization using a multiple-stimulus listening test for mixes of internalized and externalized speech stimuli presented over headphones. The results showed that HI listeners had a contracted perception of externalization correlated with high-frequency hearing loss.

Phonetic correlates of vocal attractiveness in American English.

This study explores fine‐grained phonetic vocal characteristics that underpin vocal attractiveness. In general, while it is well known that F0 plays a major role in such judgments [see, e.g., Riding et al. (2006)] there is a distinct lack of more detailed examinations of the phenomenon [see Zuta (2007) for a notable exception]. Moreover, the term “attractiveness” is generally ill‐defined and conflated with other terms (such as “pleasantness”). Therefore, the specific goal of this study is to replicate and extend such studies by including a large number of talkers, more detailed acoustic measures, and better definition the term “attractiveness”. Specifically, 60 talkers from California (30 female) produced isolated words controlled for phonetic content. These voices will be played to listeners who will judge the attractiveness of each talker. Ratings of these talkers will be compared against these acoustic measures: duration, average F0, F0 variation, spectral tilt, jitter, vowel space area, long term averaged spectrum, VOT, spectral mean of frication, and spectral peak of frication. The semantic value of “attractiveness” will be explored in follow‐up questionnaires asking more detailed questions. Results will be compared against previous studies and will be discussed in terms of possible universal and culture‐specific features.

Phoneme perception in temporally and spectrally modulated maskers: Effect of suprathreshold hearing impairment.

Normal-hearing (NH) listeners show better speech recognition in fluctuating maskers than in steady-state (SS) maskers. Hearing-impaired (HI) listeners do not show such benefit due to elevated thresholds. With amplification, HI listeners show some benefit, but significantly less than that for NH listeners. To investigate the reduced benefit in HI listeners at suprathreshold levels, we measured consonant and vowel recognition by NH and aided HI listeners in the presence of SS and modulated maskers. The modulated noises were generated by modulating the SS noise, in either one or six channels, by either one talker speech or four talker babble, resulting in four different modulated masker conditions. HI listeners were provided individualized gain according to the Cambridge formula. Both NH and aided HI listeners showed reduced intelligibility for consonants and vowels due to increase in the number of talkers. Increase in the number of channels provided more benefit to NH listeners than to HI listeners, and this benefit was greater for vowels than for consonants. This benefit due to spectral modulations for aided HI listeners was not correlated with audibility, but seems to be correlated with their ERBs. [Work supported by the Oticon Foundation, Denmark.]

Psychometric properties of the coordinate response measure corpus with various types of background interference.

The coordinate response measure (CRM) corpus has gained broad acceptance as a research tool for investigating speech intelligibility in background competition and has been widely used in studies of informational masking. Beneficial properties of the CRM include closed set format, limited linguistic content, and inclusion of target items that promote sound stream segregation based on the voice-identify of the target talker. Extensive research has demonstrated the psychometric characteristics of CRM target-word identification in the presence of one, two, or three competing talkers from the same corpus as well as long-term speech-shaped noise (LTSSN). Percent correct identification varies non-monotonically with signal-to-background ratio for one and two CRM-talker interference and is monotonic when the number of talkers exceeds 2 or is LTSSN. The CRM has even broader applicability when used with other types of background competition. The purpose of this study is to establish the psychometric characteristics of CRM target-word identification in various backgrounds with the goal of being able to determine when it is appropriate or not to use adaptive threshold procedures with the CRM corpus. Psychometric functions and thresholds based on adaptive tracking for identification of CRM target words presented in multi-talker babble and cafeteria noise will be discussed.

PRESTO: Perceptually robust English sentence test: Open-set—Design, philosophy, and preliminary findings.

Traditional clinical word recognition tests (e.g., SPIN and HINT) do not reflect variability in real world environments. These tests are often constructed with predictable short sentences spoken by a small number of talkers without dialect variation; thus, individual differences are difficult to uncover. To compensate for these shortcomings, PRESTO uses materials taken from the TIMIT database in order to incorporate variation in talkers, dialects, and number of words in a sentence. Familiarity and lexical frequency, syntactic structure, and semantic content are also considered. To provide normative data, normal-hearing young listeners performed an open-set identification task with the PRESTO materials in open field with six-talker babble mixed at different signal-to-noise ratios and also with no additional background noise. The HINT sentences were also used with the same listeners to assess the validity of PRESTO. Results indicate that listeners perform near ceiling for both the PRESTO and HINT materials in ideal listening conditions. However, under degraded listening conditions, the PRESTO shows much greater variability. Preliminary results from a clinical population also indicate that the more challenging PRESTO lists reveal individual differences among patients which are not apparent from scores on the HINT. [NIDCD T32-DC00012.]

Multitalker speech perception with ideal time-frequency segregation: Effects of voice characteristics and number of talkers

When a target voice is masked by an increasingly similar masker voice, increases in energetic masking are likely to occur due to increased spectro-temporal overlap in the competing speech waveforms. However, the impact of this increase may be obscured by informational masking effects related to the increased confusability of the target and masking utterances. In this study, the effects of target-masker similarity and the number of competing talkers on the energetic component of speech-on-speech masking were measured with an ideal time-frequency segregation (ITFS) technique that retained all the target-dominated time-frequency regions of a multitalker mixture but eliminated all the time-frequency regions dominated by the maskers. The results show that target-masker similarity has a small but systematic impact on energetic masking, with roughly a 1 dB release from masking for same-sex maskers versus same-talker maskers and roughly an additional 1 dB release from masking for different-sex masking voices. The results of a second experiment measuring ITFS performance with up to 18 interfering talkers indicate that energetic masking increased systematically with the number of competing talkers. These results suggest that energetic masking differences related to target-masker similarity have a much smaller impact on multitalker listening performance than energetic masking effects related to the number of competing talkers in the stimulus and non-energetic masking effects related to the confusability of the target and masking voices.

Subjective ratings of sentences in clear versus conversational speech.

Investigations of the relationship between speech acoustics and intelligibility face a significant methodological challenge, especially when using materials produced by multiple talkers and/or in multiple speaking styles. Ideally, acoustic analyses comparing talkers or styles will use identical materials for each. In contrast, intelligibility studies using meaningful stimuli must use either different materials or different listener groups to avoid learning effects. The latter solution becomes prohibitive when the number of talkers is large, such as in the Ferguson clear speech database (2004). While talkers recorded different materials in each speaking style, materials were the same for all talkers. For word intelligibility, familiarization prior to testing can prevent learning effects. For sentences, however, different listeners would be needed for each talker, resulting in 41 listener groups. An alternative solution is to use subjective rather than objective measures. This project explored the feasibili...

Effects of talker and token variability on perceptual learning of dialect categories.

Dialect classification is difficult for naive listeners, but perceptual learning tasks using sentence-length utterances have been shown to produce modest improvements in performance. The goal of the current study was to explore perceptual learning by naive listeners in a speeded dialect classification task with shorter, word-length utterances. In a series of experiments, participants were trained in a two-alternative forced-choice speeded dialect classification task (Cleveland vs. Cincinnati) with feedback and were then tested in the same task with novel talkers and novel words without feedback to assess learning. Variability in the stimulus materials in the training phase, including the number of talkers from each dialect and the number of different tokens produced by each talker, was manipulated across experiments to determine how variation in the input affected perceptual learning of dialect categories. The results revealed that training materials consisting of utterances produced by multiple different talkers from each dialect with multiple different tokens produced by each talker led to a significant improvement in dialect classification performance compared to a baseline condition without training. These findings suggest that dialect classification performance can improve with training on short, word-length utterances, but that robust dialect category learning requires high variability stimulus materials.

The role of the cochlear processing in human speech recognition

Little is know about how the auditory system decodes speech. We may think of speech communication re Shannon's source‐channel model, thus viewed, the most complex part of the speech communication channel is the auditory system (the receiver). In my speech‐perception research, I have attempted to limit the assumptions, and have thus fallen back on Shannon's basic source‐channel model. The basic tool is the confusion matrix (CM) for isolated natural consonant and vowels (CV), as a function of the speech to noise ratio (SNR), with several types of masking noise. We have used large numbers of talkers and listeners (i.e., 20). In a second experiment we selectively remove islands of speech in time‐frequency, and then correlate the resulting modified speech against subject scores. Our most important conclusions are: (1) The across‐frequency onset transient portion of the signal is typically the most important. (2) The spectral regions of these transient are used to code different consonants. (3) While the freque...

Effect of room absorption on human vocal output in multitalker situations

People increase their vocal output in noisy environments. This is known as the Lombard effect. The aim of the present study was to measure the effect as a function of the absorption coefficient. The noise source was generated by using other talkers in the room. A-weighted sound levels were measured in a 108 m(3) test room. The number of talkers varied from one to four and the absorption coefficients from 0.12 to 0.64. A model was introduced based on the logarithmic sum of the level found in an anechoic room plus the increasing portion of noise levels up to 80 dB. Results show that the model fits the measurements when a maximum slope of 0.5 dB per 1.0 dB increase in background level is used. Hence Lombard slopes vary from 0.2 dBdB at 50 dB background level to 0.5 dBdB at 80 dB. In addition, both measurements and the model predict a decrease of 5.5 dB per doubling of absorbing area in a room when the number of talkers is constant. Sound pressure levels increase for a doubling of talkers from 3 dB for low densities to 6 dB for dense crowds. Finally, there was correspondence between the model estimation and previous measurements reported in the literature.

Phonetic and lexical interferences in informational masking during speech-in-speech comprehension

This study investigates masking effects occurring during speech comprehension in the presence of concurrent speech signals. We examined the differential effects of acoustic–phonetic and lexical content of 4- to 8-talker babble (natural speech) or babble-like noise (reversed speech) on word identification. Behavioral results show a monotonic decrease in speech comprehension rates with an increasing number of simultaneous talkers in the reversed condition. Similar results are obtained with natural speech except for the 4-talker babble situations. An original signal analysis is then proposed to evaluate the spectro-temporal saturation of composite multitalker babble. Results from this analysis show a monotonic increase in spectro-temporal saturation with an increasing number of simultaneous talkers, for both natural and reversed speech. This suggests that informational masking consists of at least acoustic–phonetic masking which is fairly similar in the reversed and natural conditions and lexical masking which is present only with natural babble. Both effects depend on the number of talkers in the background babble. In particular, results confirm that lexical masking occurs only when some words in the babble are detectable, i.e. for a low number of talkers, such as 4, and diminishes with more talkers. These results suggest that different levels of linguistic information can be extracted from background babble and cause different types of linguistic competition for target-word identification. The use of this paradigm by psycholinguists could be of primary interest in detailing the various information types competing during lexical access.

Measurement and prediction of speech and noise levels and the Lombard effect in eating establishments

Measurements made of the acoustical characteristics of, and occupied noise levels in, ten eating establishments are described. Levels to which diners and employees were exposed varied from 45 to 82 dB(A). From these levels and diner questionnaire responses, the number of customers present and average noise levels to which individual diners were exposed during their visits were estimated. These data, assumptions about the number of talkers per customer, and classical room-acoustical theory were used to deduce talker voice output levels. These varied from slightly above "casual" to "loud." An iterative model for predicting speech and noise levels in eating establishments, including the Lombard effect as described by a new, proposed model, was developed. With the measured noise levels as the target for prediction, optimization techniques were used to find best estimates of unknown prediction parameters--such as those defining the Lombard effect, the number of talkers per customer, and the average absorption per customer--with highly credible results. The prediction algorithm and optimal parameters constitute a novel model for predicting speech and noise levels--and thus speech intelligibility--in eating establishments, as a function of the number of customers, including a proven, realistic model of the Lombard effect.

Effect of peripheral interference and attentional loss on divided listening to multiple simultaneous talkers

Effects of two factors, peripheral interference (masking) and attentional loss, that may impose restriction on divided listening were investigated using Japanese words. In each trial in the divided listening task, simultaneous multiple voices were presented first, then followed by a target voice produced by one speaker. Participants reported whether the target was included in the multiple voices or not. When the number of talkers (manipulated from one to six) was more than three, the estimated hit ratio was lower than unity, and decreased monotonically as the number of talkers increased. To investigate factors inducing the decrement, participants were also tested when the target was presented before the multiple voices. In this condition, decrement of the performance as a function of the number of talkers was gradual, compared to that in the divided listening condition. Decrement of the performance in the pretarget condition was thought to come mainly from peripheral interference, and the difference between the two tested conditions was interpreted as representing the cost in dividing attention. The result suggests that failure in divided listening to three or four talkers largely comes from the attentional factor, although the role of peripheral interference becomes larger when there are more talkers.