Speaker Characteristics Research Articles

КОМУНІКАТИВНІ ПЕРЕПИТУВАННЯ В УКРАЇНСЬКОМУ ДИСКУРСІ

This article researches communicative echo-questions in Ukrainian discourse. A communicative echo-question is a unique linguistic phenomenon which is used in the dialogue speech and has different pragmatic meanings and gives the speech communicative features. Communicative echo-questions are classified according to the pragmatic, structural and semantic aspects. They need the future research in Ukrainian discourse. Communicative echo-questions are a specific type of interrogative sentences, which repeat the previous phrase in the interrogative form and can follow different intentions. The research shows that communicative echo-questions can express indirect meanings of the communication and help resolve a conflict situation between speakers. In the research communicative echo-questions are identified on the basis of communicative relations between speakers. They can be used in different communicative situations of microdialogues. Communicative echoquestions can express politeness, rudeness, surprise and etc. The are divided into the following types, such as mental echo-questions, echo-questions which express surprise and echo-questions which are used for communicative pause. Echo-questions can be used by different groups of people, i.e. by men and women, old and young people, polite and rude persons etc. Echo-questions can be used in their speech to show their different inner state. It helps them to communicate positively avoiding conflict situations. Echo-questions are very interesting phenomenon which helps speakers to be polite with each other and communicate deeply. Social characteristics of echo-questions are also very important because they show different social features of situations and speakers with various status roles. Echo-questions help speakers to understand each other better and makes Ukrainian discourse more diplomatic. Pragmatic characteristics of communicative echo-questions show their functional role in Ukrainian discourse.

Emotional Voice Conversion Using a Hybrid Framework With Speaker-Adaptive DNN and Particle-Swarm-Optimized Neural Network

We propose a hybrid network-based learning framework for speaker-adaptive vocal emotion conversion, tested on three different datasets (languages), namely, EmoDB (German), IITKGP (Telugu), and SAVEE (English). The optimized learning model introduced is unique because of its ability to synthesize emotional speech with an acceptable perceptive quality while preserving speaker characteristics. The multilingual model is extremely beneficial in scenarios wherein emotional training data from a specific target speaker are sparsely available. The proposed model uses speaker-normalized mel-generalized cepstral coefficients for spectral training with data adaptation using the seed data from the target speaker. The fundamental frequency (F0) is transformed using a wavelet synchrosqueezed transform prior to mapping to obtain a sharpened time-frequency representation. Moreover, a feedforward artificial neural network, together with particle swarm optimization, was used for F0 training. Additionally, static-intensity modification was also performed for each test utterance. Using the framework, we were able to capture the spectral and pitch contour variabilities of emotional expression better than with other state-of-the-art methods used in this study. Considering the overall performance scores across datasets, an average melcepstral distortion (MCD) of 4.98 and root mean square error (RMSE-F0) of 10.67 were obtained in objective evaluations, and an average comparative mean opinion score (CMOS) of 3.57 and speaker similarity score of 3.70 were obtained for the proposed framework. Particularly, the best MCD of 4.09 (EmoDB-happiness) and RMSE-F0 of 9.00 (EmoDB-anger) were obtained, along with the maximum CMOS of 3.7 and speaker similarity of 4.6, thereby highlighting the effectiveness of the hybrid network model.

Text-Independent Speaker Identification Through Feature Fusion and Deep Neural Network

Speaker identification refers to the process of recognizing human voice using artificial intelligence techniques. Speaker identification technologies are widely applied in voice authentication, security and surveillance, electronic voice eavesdropping, and identity verification. In the speaker identification process, extracting discriminative and salient features from speaker utterances is an important task to accurately identify speakers. Various features for speaker identification have been recently proposed by researchers. Most studies on speaker identification have utilized short-time features, such as perceptual linear predictive (PLP) coefficients and Mel frequency cepstral coefficients (MFCC), due to their capability to capture the repetitive nature and efficiency of signals. Various studies have shown the effectiveness of MFCC features in correctly identifying speakers. However, the performances of these features degrade on complex speech datasets, and therefore, these features fail to accurately identify speaker characteristics. To address this problem, this study proposes a novel fusion of MFCC and time-based features (MFCCT), which combines the effectiveness of MFCC and time-domain features to improve the accuracy of text-independent speaker identification (SI) systems. The extracted MFCCT features were fed as input to a deep neural network (DNN) to construct the speaker identification model. Results showed that the proposed MFCCT features coupled with DNN outperformed existing baseline MFCC and time-domain features on the LibriSpeech dataset. In addition, DNN obtained better classification results compared with five machine learning algorithms that were recently utilized in speaker recognition. Moreover, this study evaluated the effectiveness of one-level and two-level classification methods for speaker identification. The experimental results showed that two-level classification presented better results than one-level classification. The proposed features and classification model for identifying a speaker can be widely applied to different types of speaker datasets.

Robust Speaker Recognition Based on Single-Channel and Multi-Channel Speech Enhancement

Deep neural network (DNN) embeddings for speaker recognition have recently attracted much attention. Compared to i-vectors, they are more robust to noise and room reverberation as DNNs leverage large-scale training. This article addresses the question of whether speech enhancement approaches are still useful when DNN embeddings are used for speaker recognition. We investigate single- and multi-channel speech enhancement for text-independent speaker verification based on x-vectors in conditions where strong diffuse noise and reverberation are both present. Single-channel (monaural) speech enhancement is based on complex spectral mapping and is applied to individual microphones. We use masking-based minimum variance distortion-less response (MVDR) beamformer and its rank-1 approximation for multi-channel speech enhancement. We propose a novel method of deriving time-frequency masks from the estimated complex spectrogram. In addition, we investigate gammatone frequency cepstral coefficients (GFCCs) as robust speaker features. Systematic evaluations and comparisons on the NIST SRE 2010 retransmitted corpus show that both monaural and multi-channel speech enhancement significantly outperform x-vector's performance, and our covariance matrix estimate is effective for the MVDR beamformer.

Prosody: A feature of languages or a feature of speakers?

Abstract Social situations of language coexistence have resulted in linguistic manifestations of bilingualism and diglossia, including linguistic interference, lexical loans and code switching. What role does prosody play in social bilingualism? In other words, when contact between different languages is not restricted to the individual but affects an entire speech community, does a dominant prosody exist? Does prosody vary among different linguistic varieties? In order to find an answer to these questions, we hereby show the results of a research project on the prosodic features of Asturian and Castilian spoken in the centre of Asturias. This experimental study is based on the speech of four informants from Oviedo – two men and two women – two of which speak Castilian, while the other two speak Asturian.

A Stylistic Analysis of Donald Trump’s Inaugural Address

Inaugural address in an important part of inauguration ceremony and every new president conceives the speech as an opportunity to promote his policies in order to appealing for people’s support. A common ground is that all presidents are fully prepared for the inaugural address and every speech has become a classic work. Donald Trump, as the new president of America, are facing changes coming from the international environment, the class consolidation, the economic downturn, and the intensification of contradictions. His inaugural address is full of his personal temperament. Although numerous researchers such as Wang Zuoliang, Ding Wangdao and Xu Zhenzhong have analyzed the stylistic features of many public speakers from varied perspectives. But there are few analyses about Trump’s speech style. And the author intends to analyze the style of his address from the aspects of phonetic feature, vocabulary feature, rhetorical feature and syntax feature. According to the author’s findings, Donald Trump raised their people’s awareness of the status quo and patriotism with the help of his unique style of speech. This paper can help us better understand the stylistic feature of speech and the connotations of Trump’s inaugural address.

A deep learning approach for speaker recognition

Speaker verification (SV) is an important branch in speaker recognition. Several approaches have been investigated within the last few decades. In this context, deep learning has received much more interest by speech processing researchers, and it was introduced recently in speaker recognition. In most cases, deep learning models are adapted from speech recognition applications and applied to speaker recognition, and they have been showing their capability of being competitors to the state-of-the-art approaches. Nevertheless, the use of deep learning in speaker recognition is still linked to speech recognition. In this study, we are proposing a new way to use deep neural networks (DNNs) in speaker recognition, in the purpose to facilitate to DNN to learn features distribution. We have been motivated by our previous work, where we have proposed a novel scoring method that works perfectly with clean speech, but it needs improvements under noisy conditions. For this reason, we are aiming to transform the extracted feature vectors (MFCCs) into enhanced feature vectors, that we denote Deep Speaker Features (DeepSFs). Experiments have been conducted on THUYG-20 SRE corpus, and significant results have been achieved. Moreover, this new method outperformed both i-vector/PLDA and our baseline system in both clean and noisy conditions.

Physical and Visual Characteristics of the Neck Predicting Gender Perception.

The necks of 30 adult men and women were measured physically and then photographed and evaluated visually by a group of 10 judges. These judges also evaluated voice recordings of the same speakers. Another group of 124 judges rated the visual and auditory masculinity/femininity of the necks and the voices. While most physical measures of the neck were larger for men, neck-length did not significantly differ between genders. A stepwise multiple regression model revealed that the single physical measure that consistently differed between genders was neck-girth (P < 0.0001). The single visual-appearance measure that consistently differed between genders was thyroid-protrusion (P = 0.0003). Neck-girth was the only physical characteristic that significantly correlated with gender differences in voice. Furthermore, the size of the thyroid prominence (ie, Adam's apple) was not associated with gender differences in voice. Neck characteristics (both physical and visual) are significantly associated with the perception of gender. While larger necks are typically perceived as masculine, neck-length is neither associated with gender nor with the speaker's voice characteristics. These findings highlight the importance of examining various physical and visual characteristics of the neck, when considering a feminization confirmation procedure for transgender patients.

Research on acquisition and recognition of Naxi speaker's speech information

The human–machine dialogue system allows people and the machine to communicate basically. This technology is also attracting more and more attention. It is conducive to the realisation of resource sharing, inheritance, and protection of Dongba culture worldwide to analyse the Naxi speech recognition for the protection and inheritance needs of Naxi Dongba culture. The model was established for Naxi speaker recognition, and the method is studied with text-independent speaker speech voice recognition. The practical analysis method was used to compare the application effect of Gaussian mixture model (GMM) and hidden Markov model (HMM) in the processing of speaker recognition in Naxi speaker's pronunciation. It is very useful to provide a theoretical basis for the establishment of a Naxi language-based human–computer interaction system to find out the speaker's voice characteristics, personality differences, and different regions. This is a theoretically necessary analysis of ethnic minority languages and the protection of ethnic and folk cultures. It can also provide practical significance for carrying out ethnic cultural differences analysis.

Introducing phonetic information to speaker embedding for speaker verification

Phonetic information is one of the most essential components of a speech signal, playing an important role for many speech processing tasks. However, it is difficult to integrate phonetic information into speaker verification systems since it occurs primarily at the frame level while speaker characteristics typically reside at the segment level. In deep neural network-based speaker verification, existing methods only apply phonetic information to the frame-wise trained speaker embeddings. To improve this weakness, this paper proposes phonetic adaptation and hybrid multi-task learning and further combines these into c-vector and simplified c-vector architectures. Experiments on National Institute of Standards and Technology (NIST) speaker recognition evaluation (SRE) 2010 show that the four proposed speaker embeddings achieve better performance than the baseline. The c-vector system performs the best, providing over 30% and 15% relative improvements in equal error rate (EER) for the core-extended and 10 s–10 s conditions, respectively. On the NIST SRE 2016, 2018, and VoxCeleb datasets, the proposed c-vector approach improves the performance even when there is a language mismatch within the training sets or between the training and evaluation sets. Extensive experimental results demonstrate the effectiveness and robustness of the proposed methods.

DNN-Based Cross-Lingual Voice Conversion Using Bottleneck Features

Cross-lingual voice conversion (CLVC) is a quite challenging task since the source and target speakers speak different languages. This paper proposes a CLVC framework based on bottleneck features and deep neural network (DNN). In the proposed method, the bottleneck features extracted from a deep auto-encoder (DAE) are used to represent speaker-independent features of speech signals from different languages. A DNN model is trained to learn the mapping between bottleneck features and the corresponding spectral features of the target speaker. The proposed method can capture speaker-specific characteristics of a target speaker, and hence requires no speech data from source speaker during training. The performance of the proposed method is evaluated using data from three Indian languages: Telugu, Tamil and Malayalam. The experimental results show that the proposed method outperforms the baseline Gaussian mixture model (GMM)-based CLVC approach.

Listener Proficiency and Shared Background Effects on the Accentedness, Comprehensibility and Intelligibility of Four Varieties of English

This study examines the impact of Hong Kong listeners’ English language proficiency on the intelligibility and perceived accentedness and comprehensibility of speakers of English from Hong Kong, China, Singapore and the United States. The study had two main aims: (1) to examine how proficiency impacts listeners’ perceptions of how accented and comprehensible different varieties of English are and how this differs from speech intelligibility; (2) to examine whether listeners benefited from a shared background effect differently by proficiency level. The research findings have pedagogical implications as they can improve understanding of which proficiency levels may benefit most from instruction and how a shared background may mitigate proficiency effects. They also help researchers understand the extent to which listeners’ own English proficiency impacts their evaluations of the speech characteristics of other speakers of English, an area of research that is still relatively unexplored.

Male fashionistas and female football fans: Gender stereotypes affect neurophysiological correlates of semantic processing during speech comprehension

Abstract Recent studies have shown that pre-existing contextual information, such as gender stereotypes, is incorporated online during comprehension (e.g., Van Berkum, van den Brink, Tesink, Kos, & Hagoort, 2008). Stereotypes, however, are not static entities, and social role theory suggests that they may be influenced by the behavior of members of the group (Eagly, 1987). Consequently, our study examines how gender stereotypes affect the semantic processing of statements from both a male and a female speaker, as well as investigating how the influence of stereotypes may change as listeners gain experience with individual speakers. Participants listened to male and female speakers produce sentences about stereotypically feminine (fashion) and stereotypically masculine (sports) topics. Half of the participants heard a stereotype congruent pattern of sentences (e.g., for the male speaker, semantic errors about fashion but no semantic errors on sports sentences) and the other half heard a stereotype incongruent pattern. We found that the N400 effect of semantic correctness is larger in stereotype incongruent conditions. Furthermore, in stereotype congruent conditions, only stimuli presented in the male voice show an N400 effect in the expected direction (larger N400s to semantic violations). Additionally, when we examined ERP changes over the course of the experiment, we found that the degree of change in amplitude was predicted by individual differences in ambivalent sexism. These results suggest that not only are speaker characteristics incorporated during online language processing, but also that social knowledge influences language processing in a manner congruent with social role theory.

Neural Predictive Coding Using Convolutional Neural Networks Toward Unsupervised Learning of Speaker Characteristics

Learning speaker-specific features is vital in many applications like speaker recognition, diarization and speech recognition. This paper provides a novel approach, we term Neural Predictive Coding (NPC), to learn speaker-specific characteristics in a completely unsupervised manner from large amounts of unlabeled training data that even contain many non-speech events and multi-speaker audio streams. The NPC framework exploits the proposed short-term active-speaker stationarity hypothesis which assumes two temporally-close short speech segments belong to the same speaker, and thus a common representation that can encode the commonalities of both the segments, should capture the vocal characteristics of that speaker. We train a convolutional deep siamese network to produce "speaker embeddings" by learning to separate `same' vs `different' speaker pairs which are generated from an unlabeled data of audio streams. Two sets of experiments are done in different scenarios to evaluate the strength of NPC embeddings and compare with state-of-the-art in-domain supervised methods. First, two speaker identification experiments with different context lengths are performed in a scenario with comparatively limited within-speaker channel variability. NPC embeddings are found to perform the best at short duration experiment, and they provide complementary information to i-vectors for full utterance experiments. Second, a large scale speaker verification task having a wide range of within-speaker channel variability is adopted as an upper-bound experiment where comparisons are drawn with in-domain supervised methods.

Structural components of readiness of students of non-linguistic faculties for the realization of intercultural communication

The article deals with the problem of readiness of students of non-linguistic specialties for the realization of intercultural communication in their future professional activity. We are talking about the competitiveness of graduates of non-linguistic universities in the conditions of the modern labor market. Relying on the fact that the notion of “readiness of a person for an activity” is defined by the author not as a temporary state, but as a stable property of a person acting constantly, the author distinguishes three structural components of the readiness of students of non-inguistic faculties to realizе intercultural communication - cognitive (knowledge about the real picture of the world, about the linguistic and cultural picture of the world), technological (ability to communicate in accordance with the cultural characteristics of the native speakers) and reflexive-evaluative (students’ skills to build and analyze the achieved results in the realization of intercultural communication and the motives). In order to form these structural components of readiness, three blocks of student training are identified: content block, technological and resulting blocks. The content block includes the knowledge of the correspondences and differences of cultural and linguistic-cultural facts, traditions, customs, customs of the native culture and the culture of the country of the language being studied. The technological block includes the forms and methods of teaching students of non- linguistic faculties how to realizе intercultural communication. The resulting block contains an evaluation of students’ knowledge and skills by the teacher. The completeness of the cognitive component of student readiness (knowledge of an intercultural nature) was determined by the author using a questionnaire. The completeness of technological and reflexive-evaluative components of the readiness of students of non-linguistic faculties to realizе intercultural communication was determined by assessing the students’ ability to perform intercultural tasks.

Spectrogram Enhanced Pitch Period Tracking using MWSG Filter in Noisy Environments

Speech Processing is the study of speech signals which carry individual information such as speaker characteristics, acoustic environment, etc due to which the parameters defining the signal are unique. Pitch Period, Duration, Intensity are the parameters that play the main role in coding speech applications such as authentication, surveillance, speaker recognition. As the conventional filters are static in nature, for non-linear and non-stationary variations of signal parameters adaptive filtering models which are robust are required. Hence the tracking and estimation of the parameters can be done by using Particle-Kalman Filter. It is very important that the signal has to track perfectly even in the presence of noise, by removing the noise and thereby enhancing the output. The approach in this paper is to propose a method for enhancing the performance, using multiple window Savitzky-Golay Filter (MWSG Filter). The performance of filter is measured by parameters Viz., SNR and PSNR.

When is irony influenced by communicative constraints? ERP evidence supporting interactive models.

Distinct theoretical proposals have described how communicative constraints (contextual biases, speaker identity) impact verbal irony processing. Modular models assume that social and contextual factors have an effect at a late stage of processing. Interactive models claim that contextual biases are considered early on. The constraint-satisfaction model further assumes that speaker's and context's characteristics can compete at early stages of analysis. The present ERP study teased apart these models by testing the impact of context and speaker features (i.e., speaker accent) on irony analysis. Spanish native speakers were presented with Spanish utterances that were ironic or literal. Each sentence was preceded by a negative or a positive context. Each story was uttered in a native or a foreign accent. Results showed that contextual biases and speaker accent interacted as early as 150ms during irony processing. Greater N400-like effects were reported for ironic than literal sentences only with positive contexts and native accent, possibly suggesting semantic difficulties when non-prototypical irony was produced by natives. A P600 effect of irony was also reported indicating inferential processing costs. The results support the constraint-satisfaction model and suggest that multiple sources of information are weighted and can interact from the earliest stages of irony analysis.

Comparison of two vibro-acoustic inverse methods to radiate a uniform sound field from a plate

The radiated sound from a conventional panel speaker, often excited by an actuator, is usually distorted due to the multi-modal response of the regular panel. To avoid such an acoustically disadvantageous feature of the conventional panel speakers, an array of actuators can be used for controlling the panel vibration to obtain the desired acoustic response. In this paper, two different control methods by using the actuator array are compared in the viewpoint of the solution stability, the input efficiency, and the acoustic performance, in particular at low frequencies rendered in a hemispherical uniform radiation. The indirect inverse rendering method is to create a vibration pattern composed of a virtual speaker and baffle, and the direct inverse rendering method controls the array actuators to form a spatial distribution of panel vibration to directly produce the uniform radiation field. The actuator positions are confined at the periphery of the plate, which are advantageous in exciting the multiple modes and saving the central zone covered by the plate for other purposes. Numerical simulations are conducted for a thin rectangular plate with simply supported boundary condition in the frequency range below 300 Hz. It is found that both rendering methods can produce the uniform hemispherical sound radiation field, but the rendering error is −20 dB for indirect method and −50 dB for direct method. The singularity problem due to the ill-conditioning of the system appears in the direct method. Effect of applying the regularization technique to obtain a stable solution is studied for the system contaminated with a large amount of noise. It is observed that the result of using the direct method becomes dramatically stable with a reduction of condition number from O (1017) to O (104). Also, the input gains of the actuator are reduced as much as 35 dB for both rendering methods.

TEnet: target speaker extraction network with accumulated speaker embedding for automatic speech recognition

It is challenging to perform automatic speech recognition when multiple people talk simultaneously. To solve this problem, speaker-aware selective methods have been proposed to extract the speech of the target speaker, relying on the auxiliary speaker characteristics provided by an anchor (a clean audio sample of the target speaker). However, the extraction performance depends on the duration and quality of the anchors, which is unstable. To address this limitation, the authors propose a target speaker extraction network (TEnet) which applies the robust speaker embedding to extract the target speech from the speech mixture. To get more stable speaker characteristics during training, the robust speaker embeddings are accumulated over all the speech of each target speaker, rather than utilising the embedding produced by a single anchor. As for testing, very few anchors are enough to get decent extraction performance. Results show the TEnet trained with accumulated embedding achieves better performance and robustness compared with the single-anchored TEnet. Moreover, to exploit the potential of the speaker embedding, the authors propose to feed the extracted target speech as anchor and train a feedback TEnet, whose results are superior to the short-anchored baseline for 22.5% on word error rate and 15.5% on signal-to-distortion rate.

Unsupervised speaker adaptation for speaker independent acoustic to articulatory speech inversion.

Speech inversion is a well-known ill-posed problem and addition of speaker differences typically makes it even harder. Normalizing the speaker differences is essential to effectively using multi-speaker articulatory data for training a speaker independent speech inversion system. This paper explores a vocal tract length normalization (VTLN) technique to transform the acoustic features of different speakers to a target speaker acoustic space such that speaker specific details are minimized. The speaker normalized features are then used to train a deep feed-forward neural network based speech inversion system. The acoustic features are parameterized as time-contextualized mel-frequency cepstral coefficients. The articulatory features are represented by six tract-variable (TV) trajectories, which are relatively speaker invariant compared to flesh point data. Experiments are performed with ten speakers from the University of Wisconsin X-ray microbeam database. Results show that the proposed speaker normalization approach provides an 8.15% relative improvement in correlation between actual and estimated TVs as compared to the system where speaker normalization was not performed. To determine the efficacy of the method across datasets, cross speaker evaluations were performed across speakers from the Multichannel Articulatory-TIMIT and EMA-IEEE datasets. Results prove that the VTLN approach provides improvement in performance even across datasets.