Speaker Characteristics Research Articles

Speaker Recognition System Using Hybrid of MFCC and RCNN with HCO Algorithm Optimization

Though there are advancements in speaker recognition technology, available systems often fail to correctly recognize speakers especially in noisy environments. The use of Mel-frequency cepstral coefficients (MFCC) has been improved using Convolutional Neural Networks (CNN) yet difficulties in achieving high accuracies still exists. Hybrid algorithms combining MFCC and Region-based Convolutional Neural Networks (RCNN) have been found to be promising. In this research features from speech signals were extracted for speaker recognition, to denoise the signals, design and develop a DFT-based denoising system using spectrum subtraction and to develop a speaker recognition method for the Verbatim Transcription using MFCC. The DFT was used to transform the sampled audio signal waveform into a frequency-domain signal. RCNN was used to model the characteristics of speakers based on their voice samples, and to classify them into different categories or identities. The novelty of the research was that it used MFCC integrated with RCNN and optimized with Host-Cuckoo Optimization (HCO) algorithm. HCO algorithm is capable of further weight optimization through the process of generating fit cuckoos for best weights. It also captured the temporal dependencies and long-term information. The system was tested and validated on audio recordings from different personalities from the National Assembly of Kenya. The results were compared with the actual identity of the speakers to confirm accuracy. The performance of the proposed approach was compared with two other existing speaker recognition the traditional approaches being MFCC-CNN and Linear Predictive Coefficients (LPC)-CNN. The comparison was based the Equal Error Rate (EER), False Rejection Rate (FRR), False Match Rate (FMR), and True Match Rate (TMR). Results show that the proposed algorithm outperformed the others in maintaining a lowest EER, FMR, FRR and highest TMR.

SUBDIALECTAL DISCOURSE AND PECULIARITIES OF ITS MENTAL CONTINUUM IN THE LENS OF THE MODERN LINGUISTIC PARADIGM

The article is devoted to the study of specific features of subdialectal discourse and its mental continuum, viewed through the prism of modern linguistic theories. The purpose of the research is to study and describe the functioning and structure features, as well as the specifics of the linguarium of subdialectal discourse. The object of the study is the subdialectal discourse, which covers speech practices recorded in the village of Pishchanyi Brid, Novoukrainskyi district, Kirovohrad region. The subject of the study is the mental continuum of the subdialect speakers, which includes knowledge, perceptions and assessments related to various aspects of their lives. The source of the study was the speech recordings of the native speakers, i.e. the textual corpus of the subdialect. The methods used in the study were modelling, hypothetical method, deductive and inductive methods, descriptive method, observation method, taxonomic method, and continuous sampling method. The peculiarity of the study of subdialectal speech in discourse studies lies in the analysis of the composition of usages, the specifics of the composition of the linguarium and its influence on the structure of discourse, expression of thoughts, interaction of speakers and the formation of the ethnographic identity of peasants who speak the subdialect. The discourse approach allows us to better understand the speech process in a specific socio-cultural and geographical setting, as well as the interaction of speakers within the dialect environment. It involves analysing the discourse of the language group in question, identifying specific lexical, grammatical and phonetic features that are characteristic of certain socio-cultural groups. The metalinguistic interpretation of discourse types in the linguistic paradigm presents dialectal discourse as a reflection of real language use, a set of idiolects and an indicator of changes in the language. The household, ceremonial, religious and existential knowledge contained in the dialect discourse is critical to the well-being and cultural integrity of rural communities. In particular, household knowledge encompasses practical skills in housekeeping, agriculture and crafts. Ceremonial and religious knowledge includes traditions, rituals and beliefs that support the spiritual life and cultural identity of the community, and existential knowledge is related to the understanding of life, death and moral values, helping community members navigate difficult life situations. Prospects for the study include a meta-analysis of dialectal discourse, which can be expressed in different ways in individual subdialects, the development of new methods for studying dialectal language in the digital environment, and the integration of interdisciplinary approaches for a deeper understanding of speech practices and cultural characteristics of speakers.

Exploring Hedging in Spoken Discourse: Insights from Corpus Analysis

The present research deals with the phenomenon of hedging language, mainly focusing on the context of spontaneous verbal interactions. The main aim of the study is to analyze the occurrence of hedges in English spoken discourse as captured in the Spoken British National Corpus 2014. The research process deals with the differences in the usage of approximators and shields existing among men and women across different age groups and levels of education. The research methodology involves creating six sub-corpora, measuring hedge dispersion, examining selected hedges to ascertain their hedging function within the context and calculating frequencies within each sub-corpus. The research outcomes point to the prevalence of shields in spoken discourse. Plausibility shields are commonly used while attribution shields show lower usage. Adaptors contribute to mitigating certainty in utterances, whereas rounders play a less prominent role. These findings emphasize the nuanced interplay between linguistic strategies and speaker characteristics in spoken communication and highlight the importance of cooperative principles in discourse.

Target speaker filtration by mask estimation for source speaker traceability in voice conversion

Voice Conversion (VC) can manipulate the source speaker's identity of speech signal to make it sound like some specific target speaker, which makes it harder for a human being or a speaker verification/identification system to trace the real identity of the source speaker. However, extracting features of the source speaker from converted audio is challenging since the features of the target speaker are dominant in the converted audio, which hinders the extraction of the features of the source speaker. In this paper, to extract features of the source speaker from audios processed by VC methods, a speaker filtration block is designed, which uses mask estimation to identify source speakers from manipulated speech signals by filtering out the features of the target speaker in converted audio. Extensive experiments are conducted to evaluate the effectiveness of the proposed model in tracing source speakers of audios converted by ADAIN-VC, AGAIN-VC, VQMIVC, and FREEVC. Experimental results demonstrate the effectiveness of the proposed model by comparing to competitive baselines in speaker verification/identification scenarios. Notably, it has good performance even when being applied to unknown VC methods. Furthermore, the experiments also show that, training audios generated by multiple VC methods can improve the performance on the traceability of the source speaker.

Parentheticals in spoken Indian and Sri Lankan English

AbstractWhile an increasing number of studies into the pragmatics of world Englishes indicate that sociobiographic factors—such as the speakers’ age or gender—influence pragmatic choices, most empirical investigations do not include sociobiographic information beyond said speaker characteristics. This study investigates parenthetical I assume/believe/feel/ guess/suppose/think in the spoken parts of the Indian and Sri Lankan English components of the International Corpus of English to answer the question of to what extent parenthetical function, that is, expressing either the speaker's opinion or insufficient knowledge, is influenced by structural, contextual and sociobiographic factors. Based on 1265 parentheticals, the results of multifactorial statistical analyses indicate that the speakers’ educational background and additional languages spoken at home are important predictors for the choice of parenthetical function. Therefore, the study calls for the inclusion of wide‐ranging sociobiographic factors (and combinations thereof) in the description of pragmatic speaker choices in world Englishes.

Whisper-SV: Adapting Whisper for low-data-resource speaker verification

Trained on 680,000 h of massive speech data, Whisper is a multitasking, multilingual speech foundation model demonstrating superior performance in automatic speech recognition, translation, and language identification. However, its applicability in speaker verification (SV) tasks remains unexplored, particularly in low-data-resource scenarios where labeled speaker data in specific domains are limited. To fill this gap, we propose a lightweight adaptor framework to boost SV with Whisper, namely Whisper-SV. Given that Whisper is not specifically optimized for SV tasks, we introduce a representation selection module to quantify the speaker-specific characteristics contained in each layer of Whisper and select the top-k layers with prominent discriminative speaker features. To aggregate pivotal speaker-related features while diminishing non-speaker redundancies across the selected top-k distinct layers of Whisper, we design a multi-layer aggregation module in Whisper-SV to integrate multi-layer representations into a singular, compacted representation for SV. In the multi-layer aggregation module, we employ convolutional layers with shortcut connections among different layers to refine speaker characteristics derived from multi-layer representations from Whisper. In addition, an attention aggregation layer is used to reduce non-speaker interference and amplify speaker-specific cues for SV tasks. Finally, a simple classification module is used for speaker classification. Experiments on VoxCeleb1, FFSVC, and IMSV datasets demonstrate that Whisper-SV achieves EER/minDCF of 2.22%/0.307, 6.14%/0.488, and 7.50%/0.582, respectively, showing superior performance in low-data-resource SV scenarios.

Domain adaptive dual-relaxation regression for speech emotion recognition

With the development of artificial intelligence techniques, speech emotion recognition (SER) has become a significant research problem. Traditional SER methods use the data from the same domain for training and testing. However, in reality, the data often come from different domains with various languages, acquisition methods, and speaker characteristics, etc. This domain discrepancy might lead to lower recognition performance and poorer generalization ability of the model. To tackle this issue, we present a novel domain adaptation (DA) method, called domain adaptive dual-relaxation regression (DADR), for cross-domain SER. In concrete, our approach first introduces a non-negative relaxation matrix to relax the strict zero-one label matrix of the source domain while learning a relaxed label matrix with a similar structure to the source labels. Then, the relaxed label matrix is utilized to learn a more discriminative transformation matrix. Further, we reduce the divergence between the source and target domains by constructing the similarity and dissimilarity graphs for cross-domain data. In addition, we add an ℓ2,1-norm to the transformation matrix to obtain more discriminative features. Experiments conducted on several cross-domain SER tasks show that our model achieves excellent recognition performance compared with several advanced algorithms.

Null subject pronouns in Papiá Kristang (Malacca Creole Portuguese)

This paper investigates variation in the subject pronoun expression (SPE) in Papiá Kristang (Malacca Creole Portuguese), a language without verb inflection. Previous accounts of Kristang (Baxter, 1988, 2012b; Hancock, 1969, 1973; Knowlton, 1964; Rêgo, 1942) present the language as predominantly displaying overt subject pronouns. However, the overall null subject pronoun rate in the oral corpus contemplated is substantial, registering a frequency of 45.7% (input pr. 0.32). As such, the paper focusses on the distribution of null subject pronouns and the constraints directing that distribution. The study considers the effects of priming, structural location, verb semantic class, and tense-mood-aspect (TMA), and addresses the effect of the individual speaker. It is found that while null subject pronoun use in Papiá Kristang resembles, with minor differences, many attributes of this mechanism in other languages, it displays one notable trait: TMA particles favor null subject pronoun use. Extralinguistic constraints on SPE within the Papiá Kristang speech community are seen to concern more the characteristics of individual speakers, and clusters of individuals, rather than predictors based on a pan-community age and gender classification.

A comparative analysis of the usage of Greek and Russian proverbs in everyday communication

The purpose of the research is to determine the linguistic status of proverbs and sayings in everyday speech of Greek and Russian native speakers. The results highlighted that both Greek and Russian speakers are well acquainted with a large number of proverbs and sayings (more than 50), but rarely use them in everyday life. Everyday paremiology is passed down from generation to generation and plays a key role in preserving cultural heritage and shaping the spiritual and moral image of society. The research also underlines the purpose behind the usage proverbs and sayings in everyday communication. It was concluded that most Greek speakers use proverbs and sayings to make information more understandable and convincing. At the same time, most Russian speakers use proverbs and sayings to express irony and relieve tension in communication. Thus, the purposes of using proverbs and sayings in everyday speech among these two nations are different. To conduct the study, the methods of sociological survey and comparative analysis were used. The scientific novelty of the research lies in the fact that the frequency of use of proverbs and sayings was determined, which reflects the characteristics of the speakers’ national identity. The practical results the research help to better understand the cultural characteristics of Greek and Russian native speakers, which helps improve the quality of intercultural communication. The above pilot data helped us to recognize the fact that the understanding and interpretation of proverbs indicate the way metaphor is used in everyday communication. Proverbs and sayings exemplify how humans use language to convey complex ideas and values through metaphorical structures. They help not only to communicate, but also to understand the world through the prism of culture and language, which is an important aspect of human thinking and cognitive abilities.

Speaker Competence Affects Prefrontal Theta and Occipital Alpha Power during Selective Word Learning in Preschoolers.

In the present study, we investigated the cognitive processes underlying selective word learning in preschoolers. We measured rhythmic neural activity in the theta (4-8 Hz) and alpha frequency range (7-12 Hz) in 67 four-year-olds. EEG was recorded during anticipation and encoding of novel labeling events performed by a speaker who had previously shown either competence (correct) or incompetence (incorrect) in labeling familiar objects. In both groups, children selected the target object equally often upon recall. However, children observing the incompetent speaker revealed weaker representations of novel words indicated by an increased likelihood for selecting familiar but incorrect items upon recall. Modulations in theta and alpha power suggest differential processing of novel label-object pairs depending on the speakers' competence. In the incompetent, but not the competent, speaker condition, increases in prefrontal theta power during anticipation and encoding were related to increased recall success. Findings suggest that theta power in the present study reflects cognitive control. In both conditions, occipital alpha power-indicating attentional processes-reflected familiarity with novel items, but in opposite directions. In familiar item trials, alpha power was increased observing the incompetent and decreased observing the competent speaker. Thus, both cognitive control and attention processes during word learning are differentially affected by speaker characteristics.

Wav2wav: Wave-to-Wave Voice Conversion

Voice conversion is the task of changing the speaker characteristics of input speech while preserving its linguistic content. It can be used in various areas, such as entertainment, medicine, and education. The quality of the converted speech is crucial for voice conversion algorithms to be useful in these various applications. Deep learning-based voice conversion algorithms, which have been showing promising results recently, generally consist of three modules: a feature extractor, feature converter, and vocoder. The feature extractor accepts the waveform as the input and extracts speech feature vectors for further processing. These speech feature vectors are later synthesized back into waveforms by the vocoder. The feature converter module performs the actual voice conversion; therefore, many previous studies separately focused on improving this module. These works combined the separately trained vocoder to synthesize the final waveform. Since the feature converter and the vocoder are trained independently, the output of the converter may not be compatible with the input of the vocoder, which causes performance degradation. Furthermore, most voice conversion algorithms utilize mel-spectrogram-based speech feature vectors without modification. These feature vectors have performed well in a variety of speech-processing areas but could be further optimized for voice conversion tasks. To address these problems, we propose a novel wave-to-wave (wav2wav) voice conversion method that integrates the feature extractor, the feature converter, and the vocoder into a single module and trains the system in an end-to-end manner. We evaluated the efficiency of the proposed method using the VCC2018 dataset.

Multimodal Unsupervised Domain Adaptation for Predicting Speaker Characteristics from Video

Multimodal Unsupervised Domain Adaptation for Predicting Speaker Characteristics from Video

Speaker Anonymization: Disentangling Speaker Features from Pre-Trained Speech Embeddings for Voice Conversion

Speech is a crucial source of personal information, and the risk of attackers using such information increases day by day. Speaker privacy protection is crucial, and various approaches have been proposed to hide the speaker’s identity. One approach is voice anonymization, which aims to safeguard speaker identity while maintaining speech content through techniques such as voice conversion or spectral feature alteration. The significance of voice anonymization has grown due to the necessity to protect personal information in applications such as voice assistants, authentication, and customer support. Building upon the S3PRL-VC toolkit and on pre-trained speech and speaker representation models, this paper introduces a feature disentanglement approach to improve the de-identification performance of the state-of-the-art anonymization approaches based on voice conversion. The proposed approach achieves state-of-the-art speaker de-identification and causes minimal impact on the intelligibility of the signal after conversion.

Universal and Language-Specific Connected Speech Characteristics of Bilingual Speakers With Alzheimer's Disease: Insights From Case Studies of Structurally Distinct Languages.

Connected speech analysis has been effectively utilized for the diagnosis and disease monitoring of individuals with Alzheimer's disease (AD). Existing research has been conducted mostly in monolingual English speakers with a noticeable lack of evidence from bilinguals and non-English speakers, particularly in non-European languages. Using a case study approach, we characterized connected speech profiles of two Bengali-English bilingual speakers with AD to determine the universal features of language impairments in both languages, identify language-specific differences between the languages, and explore language impairment characteristics of the participants with AD in relation to their bilingual language experience. Participants included two Bengali-English bilingual speakers with AD and a group of age-, gender-, education-, and language-matched neurologically healthy controls. Connected speech samples were collected in first language (L1; Bengali) and second language (L2; English) using a novel storytelling task (i.e., Frog, Where Are You?). These samples were analyzed using an augmented quantitative production analysis and correct information unit analyses for productivity, fluency, syntactic and morphosyntactic features, and lexical and semantic characteristics. Irrespective of the language, AD impacted speech productivity (speech rate and fluency) and semantic characteristics in both languages. Unique language-specific differences were noted on syntactic measures (reduced sentence length in Bengali), lexical distribution (fewer pronouns and absence of reduplication in Bengali), and inflectional properties (no difficulties with noun or verb inflections in Bengali). Among the two participants with AD, the individual who showed lower proficiency and usage in L2 (English) demonstrated reduced syntactic complexity and morphosyntactic richness in English. Evidence from these case studies suggests that language impairment features in AD are not universal across languages, particularly in comparison to impairments typically associated with language breakdowns in English. This study underscores the importance of establishing connected speech profiles in AD for non-English-speaking populations, especially for structurally different languages. This would in turn lead to the development of language-specific markers that can facilitate early detection of language deterioration and aid in improving diagnosis of AD in individuals belonging to underserved linguistically diverse populations. https://doi.org/10.23641/asha.25412458.

Learning linguistic diversity: Listeners have race-based linguistic expectations, but only for phonological variation.

In three artificial language experiments, we explored the rate at which adults learned associations between linguistic variation and speaker characteristics. Within each of the experiments, we observed that listeners sociolinguistic learning occurred, regardless of whether the speaker characteristic is social (race and sex/gender) or nonsocial (hat wearing), or whether they heard a phonological or morphological variant. However, we found that listener's initial expectations of what social properties were predictive of linguistic variation differed, impacting overall performance. First, participants were much more likely to assume that a phonological variant was predicted by a social property than a nonsocial property (Experiment 1). Most interestingly, participants were more likely to privilege speaker race than sex/gender, but only in the case of a phonological variant (Experiments 2 and 3). The same effect was found in both White and Black participants, though White participants were more likely to correctly articulate which speaker characteristic explained the variation, suggesting that sociolinguistic learning hinges on real-world experiences with language and social diversity. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Short Utterance Speaker Recognition Based on Speech High Frequency Information Compensation and Dynamic Feature Enhancement Methods

This work aims to further compensate for the weaknesses of feature sparsity and insufficient discriminative acoustic features in existing short-duration speaker recognition. To address this issue, we propose the Bark-scaled Gauss and the linear filter bank superposition cepstral coefficients (BGLCC), and the multidimensional central difference (MDCD) acoustic feature extracted method. The Bark-scaled Gauss filter bank focuses on low-frequency information, while linear filtering is uniformly distributed, therefore, the filter superposition can obtain more discriminative and richer acoustic features of short-duration audio signals. In addition, the multi-dimensional central difference method captures better dynamics features of speakers for improving the performance of short utterance speaker verification. Extensive experiments are conducted on short-duration text-independent speaker verification datasets generated from the VoxCeleb, SITW, and NIST SRE corpora, respectively, which contain speech samples of diverse lengths, and different scenarios. The results demonstrate that the proposed method outperforms the existing acoustic feature extraction approach by at least 10% in the test set. The ablation experiments further illustrate that our proposed approaches can achieve substantial improvement over prior methods.

The Influence of Disciplinary Variation and Speaker Characteristics on the Use of Hedges and Boosters in Zhihu Live Talks

The Influence of Disciplinary Variation and Speaker Characteristics on the Use of Hedges and Boosters in Zhihu Live Talks

Methodological conditions of teaching Russian as a foreign language and their implementation at Kazan Innovation University named after V.G. Timiryasov

The article analyzes the implementation by teachers of Kazan Innovative University named after V.G. Timiryasov of methodological conditions for teaching foreign students Russian as a foreign language, evaluates the effectiveness of the implementation of these conditions in terms of students achieving different levels of proficiency in the Russian language. The methodological recommendations of leading experts in the field of RCT on the maximum allowable number of students in a group of 8 people are substantiated. The requirement is being considered that classes in a group should be conducted by two teachers at the same time (in the weekly schedule, both teachers work in a 50%/50% ratio. This helps to overcome the effect of «getting used» to the timbre of a foreign student’s voice, the pace of speech and the phonetic features of a native Russian speaker. Each teacher activates all types of speech activity in each lesson: speaking, writing, listening, reading and learning vocabulary and grammar. The use of various non-academic forms of improving language skills in teaching foreign students allows students to actively train speaking and listening skills, expand their horizons by getting acquainted with the surrounding foreign cultural reality outside the framework of academic classes. A fundamentally important methodological point is the «re-equipment» of study groups after foreign students have passed the test for a certain level of Russian language acquisition, depending on the results demonstrated by the students. An important condition for ensuring the effectiveness of RCT training is interaction with the parents of students. Kazan Innovative University named after V.G. Timiryasov, when organizing the teaching of Russian as a foreign language to foreign citizens in a remote format, focuses in this process on ensuring «live» communication with the teacher.

Attention-based speech feature transfer between speakers.

In this study, we propose a simple yet effective method for incorporating the source speaker's characteristics in the target speaker's speech. This allows our model to generate the speech of the target speaker with the style of the source speaker. To achieve this, we focus on the attention model within the speech synthesis model, which learns various speaker features such as spectrogram, pitch, intensity, formant, pulse, and voice breaks. The model is trained separately using datasets specific to the source and target speakers. Subsequently, we replace the attention weights learned from the source speaker's dataset with the attention weights from the target speaker's model. Finally, by providing new input texts to the target model, we generate the speech of the target speaker with the styles of the source speaker. We validate the effectiveness of our model through similarity analysis utilizing five evaluation metrics and showcase real-world examples.

An Acoustic Analysis of Description and Classification of Balti Segmental (Velar Sounds) Consonants

The goal of this study is to categorize solely the consonants (velars) of the Balti language, which is spoken in Pakistan's Baltistan area of Gilgit Baltistan. The Balti language was chosen for the study because it is one of the less well-known languages in Pakistan and because there has never been any substantial research done on the language's sounds or sound system.120 distinct words from various Balti novels were chosen in order to categorize Balti sounds. Using the program PRAAT, the acoustic analysis of Balti consonants has been examined to demonstrate their physical characteristics. In order to capture samples of various Balti consonant sounds, the investigators identified 20 indigenous Balti speakers, aged 16 to 35, enrolled in State College for Boys in Model Town and University of Technology and Management, Lahore, Pakistan. Spectrophotograms representing each sound were used to display the physical characteristics of the native Balti speakers' sound samples. The Balti language has ten fricative sounds, according to acoustic examinations of consonant sounds.