Short Utterances Research Articles

An Analysis of the Short Utterance Problem for Speaker Characterization

Speaker characterization has always been conditioned by the length of the evaluated utterances. Despite performing well with large amounts of audio, significant degradations in performance are obtained when short utterances are considered. In this work we present an analysis of the short utterance problem providing an alternative point of view. From our perspective the performance in the evaluation of short utterances is highly influenced by the phonetic similarity between enrollment and test utterances. Both enrollment and test should contain similar phonemes to properly discriminate, being degraded otherwise. In this study we also interpret short utterances as incomplete long utterances where some acoustic units are either unbalanced or just missing. These missing units are responsible for the speaker representations to be unreliable. These unreliable representations are biased with respect to the reference counterparts, obtained from long utterances. These undesired shifts increase the intra-speaker variability, causing a significant loss of performance. According to our experiments, short utterances (3–60 s) can perform as accurate as if long utterances were involved by just reassuring the phonetic distributions. This analysis is determined by the current embedding extraction approach, based on the accumulation of local short-time information. Thus it is applicable to most of the state-of-the-art embeddings, including traditional i-vectors and Deep Neural Network (DNN) xvectors.

Back From the Future: Nonlinear Anticipation in Adults' and Children's Speech.

Purpose This study examines the temporal organization of vocalic anticipation in German children from 3 to 7 years of age and adults. The main objective was to test for nonlinear processes in vocalic anticipation, which may result from the interaction between lingual gestural goals for individual vowels and those for their neighbors over time. Method The technique of ultrasound imaging was employed to record tongue movement at 5 time points throughout short utterances of the form V1#CV2. Vocalic anticipation was examined with generalized additive modeling, an analytical approach allowing for the estimation of both linear and nonlinear influences on anticipatory processes. Results Both adults and children exhibit nonlinear patterns of vocalic anticipation over time with the degree and extent of vocalic anticipation varying as a function of the individual consonants and vowels assembled. However, noticeable developmental discrepancies were found with vocalic anticipation being present earlier in children's utterances at 3-5 years of age in comparison to adults and, to some extent, 7-year-old children. Conclusions A developmental transition towards more segmentally-specified coarticulatory organizations seems to occur from kindergarten to primary school to adulthood. In adults, nonlinear anticipatory patterns over time suggest a strong differentiation between the gestural goals for consecutive segments. In children, this differentiation is not yet mature: Vowels show greater prominence over time and seem activated more in phase with those of previous segments relative to adults.

Self-attention based speaker recognition using Cluster-Range Loss

Speaker recognition with short utterances is a challenging research topic in the natural language processing (NLP) field. Previous convolutional neural network (CNN) based models for speaker recognition usually utilize very deep or wide layers, resulting in many parameters and high computational cost. Besides, great training difficulty and inefficiency exist in the triplet loss, which is widely used in speaker recognition. In this work, we propose to combine the residual network (ResNet) with the self-attention mechanism to achieve better performance in text-independent speaker recognition with fewer parameters and lower computational cost. In addition, the Cluster-Range Loss based on a well-designed online exemplar mining is proposed to directly shrink the intra-class variation and to enlarge the inter-class distance. Experiments on Voxceleb dataset are conducted to verify the effectiveness of the proposed scheme. The proposed approach achieves a Top-1 accuracy of 89.1% for speaker identification by jointly training the network with the Cluster-Range Loss and softmax cross entropy loss. For speaker verification, we achieve a competitive EER of 5.5% without any heavy-tailed backend, compared with the state-of-the-art i-vector system, as well as the x-vector system.

End-to-end DNN based text-independent speaker recognition for long and short utterances

Recently several end-to-end speaker verification systems based on deep neural networks (DNNs) have been proposed. These systems have been proven to be competitive for text-dependent tasks as well as for text-independent tasks with short utterances. However, for text-independent tasks with longer utterances, end-to-end systems are still outperformed by standard i-vector + PLDA systems. In this work, we present an end-to-end speaker verification system that is initialized to mimic an i-vector + PLDA baseline. The system is then further trained in an end-to-end manner but regularized so that it does not deviate too far from the initial system. In this way we mitigate overfitting which normally limits the performance of end-to-end systems. The proposed system outperforms the i-vector + PLDA baseline on both long and short duration utterances.

Roar of a Champion: Loudness and Voice Pitch Predict Perceived Fighting Ability but Not Success in MMA Fighters

Historically, antagonistic interactions have been a crucial determinant of access to various fitness-affecting resources. In many vertebrate species, information about relative fighting ability is conveyed, among other things, by vocalization. Previous research found that men's upper-body strength can be assessed from voice. In the present study, we tested formidability perception of intimidating vocalization (roars) and a short utterance produced by amateur male MMA fighters attending the amateur European Championships in relation to their physical fitness indicators and fighting success. We also tested acoustic predictors of the perceived formidability. We found that body height, weight, and physical fitness failed to predict perceived formidability either from speech or from the roars. Similarly, there was no significant association between formidability of the roars and utterances and actual fighting success. Perceived formidability was predicted mainly by roars' and utterances' intensity and roars' harmonics-to-noise ratio and duration. Interestingly, fundamental frequency (F0) predicted formidability ratings in both roars and utterances but in an opposite manner, so that low F0 utterances but high F0 roars were rated as more formidable. Our results suggest that formidability perception is primarily driven by intensity and duration of the vocalizations.

Czy łatwiej jest rozpoznać emocję na podstawie swoistej ekspresji mimicznej czy prozodii?

<p>Istnieją intersubiektywne pozajęzykowe środki ekspresji emocji. Człowiek poznaje je poprzez uczestnictwo w interakcjach społecznych. Dzięki temu uczy się okazywania swoich emocji w sposób zrozumiały dla innych oraz sam rozpoznaje cudze emocje. Przeprowadzono dwa powiązane ze sobą badania. Pierwsze z nich miało postać internetowej ankiety, w której wzięły udział osoby dorosłe. Ich zadaniem było rozpoznanie, jaką emocję wyraża osoba przedstawiona na realistycznej fotografii. Drugie badanie to test audytywny, w którym wzięli udział studenci kierunku logopedia z audiologią. Nagrania wykorzystane w teście audytywnym zostały wykonane z udziałem dorosłych, zdrowych, niekształconych aktorsko osób. Zadaniem studentów było rozpoznanie emocji osoby mówiącej na podstawie krótkiej, pozbawionej szerszego kontekstu, wypowiedzi, której semantyka była emocjonalnie neutralna. Rozpoznawane emocje ujęto w następujące grupy: „cieszenie się”, „smucenie się”, „złoszczenie się” i „neutralność emocjonalna”. Wyniki obu badań przyniosły odpowiedzi m.in. na następujące pytania: Czy osoby dorosłe potrafią skutecznie rozpoznawać emocje wyłącznie na podstawie jednej modalności – wzrokowej lub słuchowej? Czy modalność sensoryczna ma znaczenie dla trudności w rozpoznawaniu emocji? Które emocje charakteryzują się zbliżoną ekspresją?</p>

Identity Vector Extraction Using Shared Mixture of PLDA for Short‐Time Speaker Recognition

The state-of-the-art speaker recognition system degrades performance rapidly dealing with short-time utterances. It is known to all that identity vectors (i-vectors) extracted from short utterances have large uncertainties and standard Probabilistic linear discriminant analysis (PLDA) method can not exploit this uncertainty to reduce the effect of duration variation. In this work, we use Shared mixture of PLDA (SM-PLDA) to remodel the i-vectors utilizing their uncertainties. SM-PLDA is an improved generative model with a shared intrinsic factor, and this factor can be regarded as an identity vector containing speaker indentification information. This identity vector can be modeled by PLDA. Experimental results are evaluated by both equal error rate and minimum detection cost function. The results conducted on the National institute of standards and technology (NIST) Speaker recognition evaluation (SRE) 2010 extended tasks show that the proposed method has achieved significant improvements compared with ivector/ PLDA and some other advanced methods.

Quality measures for speaker verification with short utterances

The performances of the automatic speaker verification (ASV) systems degrade due to the reduction in the amount of speech used for enrollment and verification. Combining multiple systems based on different features and classifiers considerably reduces speaker verification error rate with short utterances. This work attempts to incorporate supplementary information during the system combination process. We use quality of the estimated model parameters as supplementary information. We introduce a class of novel quality measures formulated using the zero-order sufficient statistics used during the i-vector extraction process. We have used the proposed quality measures as side information for combining ASV systems based on Gaussian mixture model–universal background model (GMM–UBM) and i-vector. The proposed methods demonstrate considerable improvement in speaker recognition performance on NIST SRE corpora, especially in short duration conditions. We have also observed improvement over existing systems based on different duration-based quality measures.

Children Probably Store Short Rather Than Frequent or Predictable Chunks: Quantitative Evidence From a Corpus Study.

One of the tasks faced by young children is the segmentation of a continuous stream of speech into discrete linguistic units. Early in development, syllables emerge as perceptual primitives, and the wholesale storage of syllable chunks is one possible strategy for bootstrapping the segmentation process. Here, we investigate what types of chunks children store. Our method involves selecting syllabified utterances from corpora of child-directed speech, which we vary according to (a) their length in syllables, (b) the mutual predictability of their syllables, and (c) their frequency. We then use the number of utterances within which words are contained to predict the time course of word learning, arguing that utterances which perform well at this task are also more likely to be stored, by young children, as undersegmented chunks. Our results show that short utterances are best-suited for predicting when children acquire the words contained within them, although the effect is rather small. Beyond this, we also find that short utterances are the most likely to correspond to words. Together, the two findings suggest that children may not store many complete utterances as undersegmented chunks, with most of the units that children store as hypothesized words corresponding to actual words. However, dovetailing with an item-based account of language-acquisition, when children do store undersegmented chunks, these are likely to be short sequences—not frequent or internally predictable multi-word chunks. We end by discussing implications for work on formulaic multi-word sequences.

Short Utterance Based Speech Language Identification in Intelligent Vehicles With Time-Scale Modifications and Deep Bottleneck Features

Conversations in the intelligent vehicles are usually short utterance. As the durations of the short utterances are small (e.g., less than 3 s), it is difficult to learn sufficient information to distinguish the type of languages. In this paper, we propose an end-to-end short utterances based speech language identification (SLI) approach, which is especially suitable for the short utterance based language identification. This approach is implemented with a long short-term memory (LSTM) neural network, which is designed for the SLI application in intelligent vehicles. The features used for LSTM learning are generated by a transfer learning method. The bottleneck features of a deep neural network, which are obtained for a mandarin acoustic-phonetic classifier, are used for the LSTM training. In order to improve the SLD accuracy with short utterances, a phase vocoder based time-scale modification method is utilized to reduce/increase the speech rate of the test utterance. By connecting the normal, speech rate reduced, and speech rate increased utterances, we can extend the length of the test utterances such that the performance of the SLI system is improved. The experimental results on the AP17-OLR database demonstrate that the proposed method can improve the performance of SLD, especially on short utterance. The proposed SLI has robust performance under the vehicular noisy environment.

Management of velopharyngeal insufficiency: The evolution of care and the current state of the art

Velopharyngeal insufficiency (VPI) is a condition in which there is incomplete closure of the velopharyngeal valve during speech production. This results in hypernasality and/or audible nasal emission. Nasal emission can cause additional secondary characteristics, including weak or omitted consonants, compensatory articulation productions, short utterance length, and even dysphonia. Overall, this condition affects the quality and intelligibility of speech production, which can have a significant effect on the individual's communication and social interactions. This article provides a brief overview of how the management of VPI has evolved over the past 40 years (which is the extent of this author's career). In addition, the current state of the art in VPI management is discussed from this author's perspective. Finally, a pathway is suggested for the future evolution of care for patients affected by VPI.

СПОСОБЫ ОБРАЗОВАНИЯ АНТИПОСЛОВИЦ В ТУРЕЦКОМ ЯЗЫКЕ

Basing on the analysis of the Turkish anti-proverbs - short utterances of humorous nature, formed by the distortion of classical proverbs, - the article identifies the most common ways of their formation. It is noted that the main method of creating anti-proverbs is one-component lexical replacement. The paper traces the reasons for the activation of anti-proverbs in the Turkish language at the turn of the XX-XXI centuries, dictated by radical changes in the life of the country, change in the pace of economic development and vector orientation to the traditions of the West. In this connection, the author gives the characteristic of the transformation of the Turkish society traditional values, which inevitably found its literary expression in anti-proverbs.

Generalized Variability Model for Speaker Verification

In this letter, we propose a generalized variability model as an extension to the total variability model. While the total variability model employs a standard normal prior distribution in its typical setup, the proposed generalized variability model relaxes this assumption and allows the latent variable distribution to be a mixture of Gaussians. The conventional total variability model can then be viewed as a special case of this generalized version where the number of mixture components is constrained to one. This proposed model is validated in the context of speaker verification tasks on both the standard and extended NIST SRE 2010 datasets. Experimental results show that modeling the distribution of the latent variables as a mixture of Gaussians leads to a better performance under all conditions and a greater gain can be expected for speaker verification using short utterances.

Deep neural network based i-vector mapping for speaker verification using short utterances

Text-independent speaker recognition using short utterances is a highly challenging task due to the large variation and content mismatch between short utterances. I-vector and probabilistic linear discriminant analysis (PLDA) based systems have become the standard in speaker verification applications, but they are less effective with short utterances. In this paper, we first compare two state-of-the-art universal background model (UBM) training methods for i-vector modeling using full-length and short utterance evaluation tasks. The two methods are Gaussian mixture model (GMM) based (denoted I-vector_GMM) and deep neural network (DNN) based (denoted as I-vector_DNN) methods. The results indicate that the I-vector_DNN system outperforms the I-vector_GMM system under various durations (from full length to 5 s). However, the performances of both systems degrade significantly as the duration of the utterances decreases. To address this issue, we propose two novel nonlinear mapping methods which train DNN models to map the i-vectors extracted from short utterances to their corresponding long-utterance i-vectors. The mapped i-vector can restore missing information and reduce the variance of the original short-utterance i-vectors. The proposed methods both model the joint representation of short and long utterance i-vectors: the first method trains an autoencoder first using concatenated short and long utterance i-vectors and then uses the pre-trained weights to initialize a supervised regression model from the short to long version; the second method jointly trains the supervised regression model with an autoencoder reconstructing the short utterance i-vector itself. Experimental results using the NIST SRE 2010 dataset show that both methods provide significant improvement and result in a 24.51% relative improvement in Equal Error Rates (EERs) from a baseline system. In order to learn a better joint representation, we further investigate the effect of a deep encoder with residual blocks, and the results indicate that the residual network can further improve the EERs of a baseline system by up to 26.47%. Moreover, in order to improve the short i-vector mapping to its long version, an additional vector, which represents the average value of phoneme posteriors across frames, is also added to the input, and results in a 28.43% improvement. When further testing the best-validated models of SRE10 on the Speaker In The Wild (SITW) dataset, the methods result in a 23.12% improvement on arbitrary-duration (1–5 s) short-utterance conditions.

Judgements of a speaker's personality are correlated across differing content and stimulus type.

It has previously been shown that first impressions of a speaker’s personality, whether accurate or not, can be judged from short utterances of vowels and greetings, as well as from prolonged sentences and readings of complex paragraphs. From these studies, it is established that listeners’ judgements are highly consistent with one another, suggesting that different people judge personality traits in a similar fashion, with three key personality traits being related to measures of valence (associated with trustworthiness), dominance, and attractiveness. Yet, particularly in voice perception, limited research has established the reliability of such personality judgements across stimulus types of varying lengths. Here we investigate whether first impressions of trustworthiness, dominance, and attractiveness of novel speakers are related when a judgement is made on hearing both one word and one sentence from the same speaker. Secondly, we test whether what is said, thus adjusting content, influences the stability of personality ratings. 60 Scottish voices (30 females) were recorded reading two texts: one of ambiguous content and one with socially-relevant content. One word (~500 ms) and one sentence (~3000 ms) were extracted from each recording for each speaker. 181 participants (138 females) rated either male or female voices across both content conditions (ambiguous, socially-relevant) and both stimulus types (word, sentence) for one of the three personality traits (trustworthiness, dominance, attractiveness). Pearson correlations showed personality ratings between words and sentences were strongly correlated, with no significant influence of content. In short, when establishing an impression of a novel speaker, judgments of three key personality traits are highly related whether you hear one word or one sentence, irrespective of what they are saying. This finding is consistent with initial personality judgments serving as elucidators of approach or avoidance behaviour, without modulation by time or content. All data and sounds are available on OSF (osf.io/s3cxy).

Can listeners hear the difference between children with normal hearing and children with a hearing impairment?

ABSTRACTAcoustic measurements have shown that the speech of hearing-impaired (HI) children differs from that of normally hearing (NH) children, even after several years of device use. This study focuses on the perception of HI speech in comparison to NH children’s speech. The purpose of this study was to investigate whether adult listeners can identify the speech of NH and HI children. Moreover, it is studied whether listeners’ experience and the children’s length of device use play a role in that assessment. For this study, short utterances of 7 children with a cochlear implant (CI), 7 children with an acoustic hearing aid (HA) and 7 children with NH were presented to 90 listeners who were required to specify the hearing status of each speech sample. The judges had different degrees of familiarity with hearing disorders: there were 30 audiologists, 30 primary schoolteachers and 30 inexperienced listeners. The results show that the speech of children with NH and HI can reliably be identified. However, listeners do not manage to distinguish between children with CI and HA. Children with CI are increasingly identified as NH with increasing length of device use. For children with HA, there is no similar change with longer device use. Also, experienced listeners seem to display a more lenient attitude towards atypical speech, whereas inexperienced listeners are stricter and generally consider more utterances to be produced by children with HI.

New approach for short utterance speaker identification

Recent advances in the speaker recognition (SR) field showed remarkably accurate and outperforming algorithms. However, their performances drastically degrade when the sparse amount of data is available. Nowadays, recognising a speaker identity when only a small amount of speech data is involved for testing and training remains a key consideration since many real world applications often have access to only speech data having a limited duration. In this study, the authors present a new improved approach, based on new information detected from the speech signal, to improve the task of automatic speaker identification. In doing so, they highlight how the detection of the speaker dialect can be explored to address the research problem related to short utterance SR. Results obtained with the new regional system are presented which provide a comparison between this system and the state-of-the-art systems for speaker identification task.

GMM and CNN Hybrid Method for Short Utterance Speaker Recognition

During the last few years, the speaker recognition technique has been widely attractive for its extensive application in many fields, such as speech communications, domestics services, and smart terminals. As a critical method, the Gaussian mixture model (GMM) makes it possible to achieve the recognition capability that is close to the hearing ability of human in a long speech. However, the GMM is failing to recognize a short utterance speaker with a high accuracy. Aiming at solving this problem, in this paper, we propose a novel model to enhance the recognition accuracy of the short utterance speaker recognition system. Different from traditional models based on the GMM, we design a method to train a convolutional neural network to process spectrograms, which can describe speakers better. Thus, the recognition system gains the considerable accuracy as well as the reasonable convergence speed. The experiment results show that our model can help to decrease the equal error rate of the recognition from 4.9% to 2.5%.

Towards understanding speaker discrimination abilities in humans and machines for text-independent short utterances of different speech styles.

Little is known about human and machine speaker discrimination ability when utterances are very short and the speaking style is variable. This study compares text-independent speaker discrimination ability of humans and machines based on utterances shorter than 2 s in two different speaking styles (read sentences and speech directed towards pets, characterized by exaggerated prosody). Recordings of 50 female speakers drawn from the UCLA Speaker Variability Database were used as stimuli. Performance of 65 human listeners was compared to i-vector-based automatic speaker verification systems using mel-frequency cepstral coefficients, voice quality features, which were inspired by a psychoacoustic model of voice perception, or their combination by score-level fusion. Humans always outperformed machines, except in the case of style-mismatched pairs from perceptually-marked speakers. Speaker representations by humans and machines were compared using multi-dimensional scaling (MDS). Canonical correlation analysis showed a weak correlation between machine and human MDS spaces. Multiple regression showed that means of voice quality features could represent the most important human MDS dimension well, but not the dimensions from machines. These results suggest that speaker representations by humans and machines are different, and machine performance might be improved by better understanding how different acoustic features relate to perceived speaker identity.

Is infant-directed speech interesting because it is surprising? – Linking properties of IDS to statistical learning and attention at the prosodic level

The exaggerated intonation and special rhythmic properties of infant-directed speech (IDS) have been hypothesized to attract infants’ attention to the speech stream. However, there has been little work actually connecting the properties of IDS to models of attentional processing or perceptual learning. A number of such attention models suggest that surprising or novel perceptual inputs attract attention, where novelty can be operationalized as the statistical (un)predictability of the stimulus in the given context. Since prosodic patterns such as F0 contours are accessible to young infants who are also known to be adept statistical learners, the present paper investigates a hypothesis that F0 contours in IDS are less predictable than those in adult-directed speech (ADS), given previous exposure to both speaking styles, thereby potentially tapping into basic attentional mechanisms of the listeners in a similar manner that relative probabilities of other linguistic patterns are known to modulate attentional processing in infants and adults. Computational modeling analyses with naturalistic IDS and ADS speech from matched speakers and contexts show that IDS intonation has lower overall temporal predictability even when the F0 contours of both speaking styles are normalized to have equal means and variances. A closer analysis reveals that there is a tendency of IDS intonation to be less predictable at the end of short utterances, whereas ADS exhibits more stable average predictability patterns across the full extent of the utterances. The difference between IDS and ADS persists even when the proportion of IDS and ADS exposure is varied substantially, simulating different relative amounts of IDS heard in different family and cultural environments. Exposure to IDS is also found to be more efficient for predicting ADS intonation contours in new utterances than exposure to the equal amount of ADS speech. This indicates that the more variable prosodic contours of IDS also generalize to ADS, and may therefore enhance prosodic learning in infancy. Overall, the study suggests that one reason behind infant preference for IDS could be its higher information value at the prosodic level, as measured by the amount of surprisal in the F0 contours. This provides the first formal link between the properties of IDS and the models of attentional processing and statistical learning in the brain. However, this finding does not rule out the possibility that other differences between the IDS and ADS also play a role.