Speech Identification Performance Research Articles

Communication Under Sharply Degraded Auditory Input and the "2-Sentence" Problem.

Despite performing well in standard clinical assessments of speech perception, many cochlear implant (CI) users report experiencing significant difficulties when listening in real-world environments. We hypothesize that this disconnect may be related, in part, to the limited ecological validity of tests that are currently used clinically and in research laboratories. The challenges that arise from degraded auditory information provided by a CI, combined with the listener's finite cognitive resources, may lead to difficulties when processing speech material that is more demanding than the single words or single sentences that are used in clinical tests. Here, we investigate whether speech identification performance and processing effort (indexed by pupil dilation measures) are affected when CI users or normal-hearing control subjects are asked to repeat two sentences presented sequentially instead of just one sentence. Response accuracy was minimally affected in normal-hearing listeners, but CI users showed a wide range of outcomes, from no change to decrements of up to 45 percentage points. The amount of decrement was not predictable from the CI users' performance in standard clinical tests. Pupillometry measures tracked closely with task difficulty in both the CI group and the normal-hearing group, even though the latter had speech perception scores near ceiling levels for all conditions. Speech identification performance is significantly degraded in many (but not all) CI users in response to input that is only slightly more challenging than standard clinical tests; specifically, when two sentences are presented sequentially before requesting a response, instead of presenting just a single sentence at a time. This potential "2-sentence problem" represents one of the simplest possible scenarios that go beyond presentation of the single words or sentences used in most clinical tests of speech perception, and it raises the possibility that even good performers in single-sentence tests may be seriously impaired by other ecologically relevant manipulations. The present findings also raise the possibility that a clinical version of a 2-sentence test may provide actionable information for counseling and rehabilitating CI users, and for people who interact with them closely.

Enhanced Speech Emotion Recognition Using DCGAN-Based Data Augmentation

Although emotional speech recognition has received increasing emphasis in research and applications, it remains challenging due to the diversity and complexity of emotions and limited datasets. To address these limitations, we propose a novel approach utilizing DCGAN to augment data from the RAVDESS and EmoDB databases. Then, we assess the efficacy of emotion recognition using mel-spectrogram data by utilizing a model that combines CNN and BiLSTM. The preliminary experimental results reveal that the suggested technique contributes to enhancing the emotional speech identification performance. The results of this study provide directions for further development in the field of emotional speech recognition and the potential for practical applications.

Development of open-set test in Marathi language for assessing speech identification performance of native children

Development of open-set test in Marathi language for assessing speech identification performance of native children

Spectral-temporally modulated ripple test Lite for computeRless Measurement (SLRM): A Nonlinguistic Test for Audiology Clinics.

Many clinics are faced with the difficulty of evaluating performance in patients who speak a language for which there are no validated tests. It would be desirable to have a nonlinguistic method of evaluating these patients. Spectral ripple tests are nonlinguistic and highly correlated with speech identification performance. However, they are generally not amenable to clinical environments as they typically require the use of computers which are often not found in clinic sound booths. In this study, we evaluate the Spectral-temporally Modulated Ripple Test (SMRT) Lite for computeRless Measurement (SLRM), which is a new variant of the adaptive SMRT that can be implemented via a CD player. SMRT and SLRM were measured for 10 normal hearing and 10 cochlear implant participants. Performance on the two tests was highly correlated (r = 0.97). The results suggest that SLRM can be used interchangeably with SMRT but can be implemented without a computer.

Translated vs Non-Translated Method for Multilingual Hate Speech Identification in Twitter

Nowadays social media is often misused to spread hate speech. Spreading hate speech is an act that needs to be handled in a special way because it can undermine or discriminate other people and cause conflict that leading to both material and immaterial losses. There are several challenges in building a hate speech identification system; one of them is identifying hate speech in multilingual scope. In this paper, we adapt and compare two methods in multilingual text classification which are translated (with and without language identification) and non-translated method for multilingual hate speech identification (including Hindi, English, and Indonesian language) using machine learning approach. We use some classification algorithms (classifiers) namely Support Vector Machine (SVM), Naive Bayes (NB), and Random Forest Decision Tree (RFDT) with word n-grams and char n-grams (character n-grams) as feature extraction. Our experiment result shows that the non-translated method gives the best result. However, the use of non-translated method needs to be reconsidered because this method needs more cost for data collection and annotation. Meanwhile, translated without language identification method give a poor result. To address this problem, we combine translated method with monolingual hate speech identification, and the experiment result shows that this approach can increase the multilingual hate speech identification performance compared to translate without language identification. This paper discusses the advantages and disadvantages for all method and the future works to enhance the performance in multilingual hate speech identification.

Vowel and consonant confusions from spectrally manipulated stimuli designed to simulate poor cochlear implant electrode-neuron interfaces.

Suboptimal interfaces between cochlear implant (CI) electrodes and auditory neurons result in a loss or distortion of spectral information in specific frequency regions, which likely decreases CI users' speech identification performance. This study exploited speech acoustics to model regions of distorted CI frequency transmission to determine the perceptual consequences of suboptimal electrode-neuron interfaces. Normal hearing adults identified naturally spoken vowels and consonants after spectral information was manipulated through a noiseband vocoder: either (1) low-, middle-, or high-frequency regions of information were removed by zeroing the corresponding channel outputs, or (2) the same regions were distorted by splitting filter outputs to neighboring filters. These conditions simulated the detrimental effects of suboptimal CI electrode-neuron interfaces on spectral transmission. Vowel and consonant confusion patterns were analyzed with sequential information transmission, perceptual distance, and perceptual vowel space analyses. Results indicated that both types of spectral manipulation were equally destructive. Loss or distortion of frequency information produced similar effects on phoneme identification performance and confusion patterns. Consonant error patterns were consistently based on place of articulation. Vowel confusions showed that perceptions gravitated away from the degraded frequency region in a predictable manner, indicating that vowels can probe frequency-specific regions of spectral degradations.

Conventional speech identification test in Marathi for adults

<p class="abstract"><strong>Background:</strong> The present study aimed to develop conventional speech identification in Marathi for assessing adults by considering word frequency, familiarity, words in common use and phonemic balancing.</p><p class="abstract"><strong>Methods:</strong> A total of four word lists were developed with each word list consisting of 25 words out of which 60% are monosyllabic words in CVC structure, and 40% are disyllabic words in CVCV structure. Equivalence analysis and performance-intensity function testing was carried out using four word lists on native speakers of Marathi belonging to different regions of Maharashtra (i.e. Vidarbha, Marathwada, Khandesh and Northern Maharashtra, Konkan and Pune) who were equally divided into five groups based on above mentioned regions. </p><p class="abstract"><strong>Results:</strong> The results revealed that there was no statistically significant difference (p >0.05) in the speech identification performance between groups for each word list, and between word lists for each group. The performance-intensity (PI) function curve showed semi-linear function, and the groups’ mean slope of the curve indicated an average slope of 4.5% increase in speech identification score per dB for four word lists. Although, there is no data available on speech identification tests for adults in Marathi, most of the findings of the study are in line with the findings of research reports on other Indian languages.</p><p><strong>Conclusions:</strong> The four word lists developed were found to be equally difficult for all the groups and can be used interchangeably. Thus, the developed word lists were found to be reliable and valid materials for assessing speech identification performance of adults in Marathi.</p><p> </p>

Is There a Relationship Between Speech Identification in Noise and Categorical Perception in Children With Dyslexia?

Children with dyslexia have been suggested to experience deficits in both categorical perception (CP) and speech identification in noise (SIN) perception. However, results regarding both abilities are inconsistent, and the relationship between them is still unclear. Therefore, this study aimed to investigate the relationship between CP and the psychometric function of SIN perception. Sixteen children with dyslexia, 16 chronological-age controls, and 16 reading-level controls were evaluated in CP of a voicing continuum and in consonant identification in both stationary and fluctuating noises. There was a small but significant impairment in speech identification performance of children with dyslexia in stationary noise compared with chronological age-matched controls (but not reading level-matched controls). However, their performance increased in a fluctuating background, hence suggesting normal masking and unmasking effects and preserved sensory processing of speech information. Regarding CP, location of the phoneme boundary differed in the children with dyslexia compared with both control groups. However, scrutinizing individual profiles failed to reveal consistently poor performance in SIN and CP tasks. In addition, there was no significant correlation between CP, SIN perception, and reading scores in the group with dyslexia. The relationship between the SIN deficit and CP, and how they potentially affect reading in children with dyslexia, remains unclear. However, these results are inconsistent with the notion that children with dyslexia suffer from a low-level temporal processing deficit and rather suggest a role of nonsensory (e.g., attentional) factors in their speech perception difficulties.

Spectrotemporal Modulation Detection and Speech Perception by Cochlear Implant Users.

Spectrotemporal modulation (STM) detection performance was examined for cochlear implant (CI) users. The test involved discriminating between an unmodulated steady noise and a modulated stimulus. The modulated stimulus presents frequency modulation patterns that change in frequency over time. In order to examine STM detection performance for different modulation conditions, two different temporal modulation rates (5 and 10 Hz) and three different spectral modulation densities (0.5, 1.0, and 2.0 cycles/octave) were employed, producing a total 6 different STM stimulus conditions. In order to explore how electric hearing constrains STM sensitivity for CI users differently from acoustic hearing, normal-hearing (NH) and hearing-impaired (HI) listeners were also tested on the same tasks. STM detection performance was best in NH subjects, followed by HI subjects. On average, CI subjects showed poorest performance, but some CI subjects showed high levels of STM detection performance that was comparable to acoustic hearing. Significant correlations were found between STM detection performance and speech identification performance in quiet and in noise. In order to understand the relative contribution of spectral and temporal modulation cues to speech perception abilities for CI users, spectral and temporal modulation detection was performed separately and related to STM detection and speech perception performance. The results suggest that that slow spectral modulation rather than slow temporal modulation may be important for determining speech perception capabilities for CI users. Lastly, test–retest reliability for STM detection was good with no learning. The present study demonstrates that STM detection may be a useful tool to evaluate the ability of CI sound processing strategies to deliver clinically pertinent acoustic modulation information.

Release from informational masking in a monaural competing-speech task with vocoded copies of the maskers presented contralaterally.

Single-sided deafness prevents access to the binaural cues that help normal-hearing listeners extract target speech from competing voices. Little is known about how listeners with one normal-hearing ear might benefit from access to severely degraded audio signals that preserve only envelope information in the second ear. This study investigated whether vocoded masker-envelope information presented to one ear could improve performance for normal-hearing listeners in a multi-talker speech-identification task presented to the other ear. Target speech and speech or non-speech maskers were presented unprocessed to the left ear. The right ear received no signal, or either an unprocessed or eight-channel noise-vocoded copy of the maskers. Presenting the vocoded maskers contralaterally yielded significant masking release from same-gender speech maskers, albeit less than in the unprocessed case, but not from opposite-gender speech, stationary-noise, or modulated-noise maskers. Unmasking also occurred with as few as two vocoder channels and when an attenuated copy of the target signal was added to the maskers before vocoding. These data show that delivering masker-envelope information contralaterally generates masking release in situations where target-masker similarity impedes monaural speech-identification performance. By delivering speech-envelope information to a deaf ear, cochlear implants for single-sided deafness have the potential to produce a similar effect.

Optimal combination of neural temporal envelope and fine structure cues to explain speech identification in background noise.

The dichotomy between acoustic temporal envelope (ENV) and fine structure (TFS) cues has stimulated numerous studies over the past decade to understand the relative role of acoustic ENV and TFS in human speech perception. Such acoustic temporal speech cues produce distinct neural discharge patterns at the level of the auditory nerve, yet little is known about the central neural mechanisms underlying the dichotomy in speech perception between neural ENV and TFS cues. We explored the question of how the peripheral auditory system encodes neural ENV and TFS cues in steady or fluctuating background noise, and how the central auditory system combines these forms of neural information for speech identification. We sought to address this question by (1) measuring sentence identification in background noise for human subjects as a function of the degree of available acoustic TFS information and (2) examining the optimal combination of neural ENV and TFS cues to explain human speech perception performance using computational models of the peripheral auditory system and central neural observers. Speech-identification performance by human subjects decreased as the acoustic TFS information was degraded in the speech signals. The model predictions best matched human performance when a greater emphasis was placed on neural ENV coding rather than neural TFS. However, neural TFS cues were necessary to account for the full effect of background-noise modulations on human speech-identification performance.

Use of Amplitude Modulation Cues Recovered from Frequency Modulation for Cochlear Implant Users When Original Speech Cues Are Severely Degraded

Won et al. (J Acoust Soc Am 132:1113-1119, 2012) reported that cochlear implant (CI) speech processors generate amplitude-modulation (AM) cues recovered from broadband speech frequency modulation (FM) and that CI users can use these cues for speech identification in quiet. The present study was designed to extend this finding for a wide range of listening conditions, where the original speech cues were severely degraded by manipulating either the acoustic signals or the speech processor. The manipulation of the acoustic signals included the presentation of background noise, simulation of reverberation, and amplitude compression. The manipulation of the speech processor included changing the input dynamic range and the number of channels. For each of these conditions, multiple levels of speech degradation were tested. Speech identification was measured for CI users and compared for stimuli having both AM and FM information (intact condition) or FM information only (FM condition). Each manipulation degraded speech identification performance for both intact and FM conditions. Performance for the intact and FM conditions became similar for stimuli having the most severe degradations. Identification performance generally overlapped for the intact and FM conditions. Moreover, identification performance for the FM condition was better than chance performance even at the maximum level of distortion. Finally, significant correlations were found between speech identification scores for the intact and FM conditions. Altogether, these results suggest that despite poor frequency selectivity, CI users can make efficient use of AM cues recovered from speech FM in difficult listening situations.

The role of syntax in maintaining the integrity of streams of speech.

This study examined the ability of listeners to utilize syntactic structure to extract a target stream of speech from among competing sounds. Target talkers were identified by voice or location, which was held constant throughout a test utterance, and paired with correct or incorrect (random word order) target sentence syntax. Both voice and location provided reliable cues for identifying target speech even when other features varied unpredictably. The target sentences were masked either by predominantly energetic maskers (noise bursts) or by predominantly informational maskers (similar speech in random word order). When the maskers were noise bursts, target sentence syntax had relatively minor effects on identification performance. However, when the maskers were other talkers, correct target sentence syntax resulted in significantly better speech identification performance than incorrect syntax. Furthermore, conformance to correct syntax alone was sufficient to accurately identify the target speech. The results were interpreted as supporting the idea that the predictability of the elements comprising streams of speech, as manifested by syntactic structure, is an important factor in binding words together into coherent streams. Furthermore, these findings suggest that predictability is particularly important for maintaining the coherence of an auditory stream over time under conditions high in informational masking.

Simultaneous suppression of noise and reverberation in cochlear implants using a ratio masking strategy

Cochlear implant (CI) recipients' ability to identify words is reduced in noisy or reverberant environments. The speech identification task for CI users becomes even more challenging in conditions where both reverberation and noise co-exist as they mask the spectro-temporal cues of speech in a rather complementary fashion. Ideal channel selection (ICS) was found to result in significantly more intelligible speech when applied to the noisy, reverberant, as well as noisy reverberant speech. In this study, a blind single-channel ratio masking strategy is presented to simultaneously suppress the negative effects of reverberation and noise on speech identification performance for CI users. In this strategy, noise power spectrum is estimated from the non-speech segments of the utterance while reverberation spectral variance is computed as a delayed and scaled version of the reverberant speech spectrum. Based on the estimated noise and reverberation power spectra, a weight between 0 and 1 is assigned to each time-frequency unit to form the final mask. Listening experiments conducted with CI users in two reverberant conditions (T60 = 0.6 and 0.8 s) at a signal-to-noise ratio of 15 dB indicate substantial improvements in speech intelligibility in both reverberant-alone and noisy reverberant conditions considered.

Auditory Training: Evidence for Neural Plasticity in Older Adults.

Improvements in digital amplification, cochlear implants, and other innovations have extended the potential for improving hearing function; yet, there remains a need for further hearing improvement in challenging listening situations, such as when trying to understand speech in noise or when listening to music. Here, we review evidence from animal and human models of plasticity in the brain's ability to process speech and other meaningful stimuli. We considered studies targeting populations of younger through older adults, emphasizing studies that have employed randomized controlled designs and have made connections between neural and behavioral changes. Overall results indicate that the brain remains malleable through older adulthood, provided that treatment algorithms have been modified to allow for changes in learning with age. Improvements in speech-in-noise perception and cognition function accompany neural changes in auditory processing. The training-related improvements noted across studies support the need to consider auditory training strategies in the management of individuals who express concerns about hearing in difficult listening situations. Given evidence from studies engaging the brain's reward centers, future research should consider how these centers can be naturally activated during training.

Effect of fundamental-frequency and sentence-onset differences on speech-identification performance of young and older adults in a competing-talker background

This study investigated the benefits of differences between sentences in fundamental frequency (F0) and temporal onset for sentence pairs among listener groups differing in age and hearing sensitivity. Two experiments were completed with the primary difference between experiments being the way in which the stimuli were presented. Experiment 1 used blocked stimulus presentation, which ultimately provided redundant acoustic cues to mark the target sentence in each pair, whereas Experiment 2 sampled a slightly more restricted stimulus space, but in a completely randomized presentation order. For both experiments, listeners were required to detect a cue word ("Baron") for the target sentence in each pair and to then identify the target words (color, number) that appeared later in the target sentence. Results of Experiment 1 showed that F0 or onset separation cues were beneficial to both cue-word detection and color-number identification performance. There were no significant differences across groups in the ability to detect the cue word, but groups differed in their ability to identify the correct color-number words. Elderly adults with impaired hearing had the greatest difficulty with the identification task despite the application of spectral shaping to restore the audibility of the speech stimuli. For the most part, the primary results of Experiment 1 were replicated in Experiment 2, although, in the latter experiment, all older adults, whether they had normal or impaired hearing, performed worse than young adults with normal hearing. From Experiment 2, the benefits received for a difference in F0 between talkers of 6 semitones were equivalent to those received for an onset asynchrony of 300 ms between sentences and, for such conditions, the combination of both sound-segregation cues resulted in an additive benefit.

Speech perception abilities for spectrally modified signals in individuals with auditory dys-synchrony

Objective. The objective of the study was to determine the speech perception abilities for spectrally modified speech signals in individuals with auditory dys-synchrony. Study Sample. The speech identification scores of 30 normal hearing and 12 individuals diagnosed as having auditory dys-synchrony were studied. Design. Phonemically balanced words in Kannada developed by were presented unfiltered and filtered at 1700 Hz low-pass and 1700 Hz high-pass cut-off frequencies, and speech identification performance was assessed. Results. Results revealed that there is a highly significant difference in speech identification scores for unfiltered and low-pass filtered words. This difference could be attributed to the physiological coding deficits of low frequency information which are usually coded by phase locked responses in auditory nerve fibers (). Conclusions. Thus, low-pass filtered words can be used as an effective tool to identify individuals with auditory dys-synchrony, especially those who have good speech identification scores in quiet.SumarioObjetivo. El objetivo de este estudio fue determinar las habilidades de percepción del lenguaje con señales de lenguaje espectralmente modificadas en sujetos con disincronía auditiva. Muestra de Estudio. Se estudiaron las puntuaciones de identificación de lenguaje de 30 normo-oyentes y de 12 individuos con diagnóstico de disincronía auditiva. Diseño. Se presentaron palabras fonéticamente balanceadas en Kannada, desarrolladas por sin filtro y con filtro pasa/bajo de 1700 Hz y filtro pasa/alto de corte de frecuencias de 1700 Hz, y se evaluó el rendimiento en la identificación del lenguaje. Resultados. Los resultados revelan que existe una diferencia significativa muy alta en las puntuaciones de identificación de lenguaje para palabras no filtradas y con filtro pasa bajo. Esta diferencia puede ser atribuida al déficit fisiológico de codificación de la información e bajas frecuencias, que usualmente se codifican como respuestas con bloqueo de fase en las fibras del nervio auditivo (). Conclusiones. Por lo tanto, las palabras filtradas pasa bajo pueden usarse como una herramienta útil para identificar personas con disincronía auditiva, especialmente aquellas que tienen buenas puntuaciones para la identificación de lenguaje en silencio.

Changes in Sensory Evoked Responses Coincide with Rapid Improvement in Speech Identification Performance

Perceptual learning is sometimes characterized by rapid improvements in performance within the first hour of training (fast perceptual learning), which may be accompanied by changes in sensory and/or response pathways. Here, we report rapid physiological changes in the human auditory system that coincide with learning during a 1-hour test session in which participants learned to identify two consonant vowel syllables that differed in voice onset time. Within each block of trials, listeners were also presented with a broadband noise control stimulus to determine whether changes in auditory evoked potentials were specific to the trained speech cue. The ability to identify the speech sounds improved from the first to the fourth block of trials and remained relatively constant thereafter. This behavioral improvement coincided with a decrease in N1 and P2 amplitude, and these learning-related changes differed from those observed for the noise stimulus. These training-induced changes in sensory evoked responses were followed by an increased negative peak (between 275 and 330 msec) over fronto-central sites and by an increase in sustained activity over the parietal regions. Although the former was also observed for the noise stimulus, the latter was specific to the speech sounds. The results are consistent with a top-down nonspecific attention effect on neural activity during learning as well as a more learning-specific modulation, which is coincident with behavioral improvements in speech identification.

Aided speech‐identification performance in single‐talker competition by older adults with impaired hearing

This study examined the effects of increased processing load on the closed-set speech-identification performance of young and older adults in a one-talker background. Since the older adults had impaired hearing, speech-identification performance was measured for spectrally shaped stimuli comparable to those experienced when wearing well-fit hearing aids. There were three groups of listeners: (1) 19 older adults with high-frequency sensorineural hearing loss; (2) 10 young adults with normal hearing who were assessed with the same spectrally shaped stimuli as the older adults; (3) 9 young adults with normal hearing who were assessed without spectral shaping and at a poorer target-to-competition ratio in an effort to equate overall performance to that of the older adults. In addition to this group factor, there were three within-participant repeated-measures independent variables designed to increase the demands on processing for the target and competing speech stimuli. These were: (1) competition meaningfulness (played in forward or reverse direction); (2) gender match between target and competing talkers (same or different gender); and (3) talker uncertainty (either the same target/competition talker pair or one of many such pairs on each trial). These three repeated-measures independent variables were examined in a 2 x 2 x 2 factorial design. They showed roughly independent and additive effects on speech-identification such that combinations of these variables decreased performance cumulatively. Older adults performed worse than young adults across the board, but also showed diminished relative improvement as the processing load was decreased. Individual differences in performance among the older adults were also examined.

The effect of hearing impairment on the identification of speech that is modulated synchronously or asynchronously across frequency

This study investigated the effect of mild-to-moderate sensorineural hearing loss on the ability to identify speech in noise for vowel-consonant-vowel tokens that were either unprocessed, amplitude modulated synchronously across frequency, or amplitude modulated asynchronously across frequency. One goal of the study was to determine whether hearing-impaired listeners have a particular deficit in the ability to integrate asynchronous spectral information in the perception of speech. Speech tokens were presented at a high, fixed sound level and the level of a speech-shaped noise was changed adaptively to estimate the masked speech identification threshold. The performance of the hearing-impaired listeners was generally worse than that of the normal-hearing listeners, but the impaired listeners showed particularly poor performance in the synchronous modulation condition. This finding suggests that integration of asynchronous spectral information does not pose a particular difficulty for hearing-impaired listeners with mild/moderate hearing losses. Results are discussed in terms of common mechanisms that might account for poor speech identification performance of hearing-impaired listeners when either the masking noise or the speech is synchronously modulated.