Speaker Models Research Articles

Autonomy-supportive treatment for acquired apraxia of speech: feasibility and therapeutic effect

ABSTRACT Background: Most treatments for acquired apraxia of speech (AOS) rely on clinician-controlled practice conditions and repeated exposure to unimpaired speaker models. Recent motor learning research indicates that autonomy-support, expectation of competence, and external attentional focus may be more beneficial for motivation and skill learning. Aims: We evaluated the feasibility and basic therapeutic effect for the initial phase of a new speech production treatment program, ActionSC, that uses self-modeling and clinician coaching to help learners with AOS build their own practice program. Methods and Procedures: The single participant was a woman with moderate AOS and nonfluent aphasia. She met with project staff twice per week to review practice strategies, develop and adjust self-modeled video cues, work on her speech, and monitor progress. The program was structured around a custom app installed on a tablet computer. Most practice was directed by the participant based on options provided by the treating clinician. We used a multiple baseline across behavior design to evaluate the relationship between this treatment and oral reading probe performance for 30 conversational phrases the participant wanted to learn to say. Outcomes and Results: Experimental control was demonstrated, with target phrases remaining at baseline levels, then improving at the time treatment was introduced sequentially across three conversation topics. Effect sizes were moderate to large after 9–12 therapy sessions plus independent home practice. The participant assumed an active role in evaluating her own performance, administering and adjusting cues, and organizing her home practice. Conclusions: An autonomy-supportive and confidence-building format for speech practice can be feasible and effective for people with AOS. Fixed cueing hierarchies, augmented feedback, and attentional focus on speech movements may be less important in AOS treatment than previously thought. In addition to replicating our preliminary results with other participants and circumstances, there is a need to extend treatment development to later learning phases in order to promote positive change in real-life settings.

Analysis of Speaker Diarization Based on Bayesian HMM With Eigenvoice Priors

In our previous work, we introduced our Bayesian Hidden Markov Model with eigenvoice priors, which has been recently recognized as the state-of-the-art model for Speaker Diarization. In this article we present a more complete analysis of the Diarization system. The inference of the model is fully described and derivations of all update formulas are provided for a complete understanding of the algorithm. An extensive analysis on the effect, sensitivity and interactions of all model parameters is provided, which might be used as a guide for their optimal setting. The newly introduced speaker regularization coefficient allows us to control the number of speakers inferred in an utterance. A naive speaker model merging strategy is also presented, which allows to drive the variational inference out of local optima. Experiments for the different diarization scenarios are presented on CALLHOME and DIHARD datasets.

The promotion of the good intercultural speaker through intercultural contacts in a Tunisian EFL textbook

ABSTRACTThis paper evaluates the potential of intercultural contacts portrayed in the content and tasks of a Tunisian EFL textbook for third-year secondary school students to develop the model of the intercultural speaker. The quantitative and qualitative analyses of the elements of intercultural contacts reveal that this portrayal is imbalanced in terms of frequency and variety and that their potential to help participants become good intercultural speakers is low. To compete with other sources of intercultural contacts, the textbook should include more frequent and varied contact situations where culturally different characters can interact and apply intercultural skills.

Speaker Modeling Using Emotional Speech for More Robust Speaker Identification

Automatic identity recognition in fast, reliable and non-intrusive way is one of the most challenging topics in digital world of today. A possible approach to identity recognition is the identification by voice. Characteristics of speech relevant for automatic speaker recognition can be affected by external factors such as noise and channel distortions, but also by speaker-specific conditions—emotional or health states. The improvement of a speaker recognition system by different model training strategies are addressed in this paper in order to obtain the best performance of the system with only a limited amount of neutral and emotional speech data. The models adopted are a Gaussian Mixture Model and i-vectors whose inputs are Mel Frequency Cepstral Coefficients, and the experiments have been conducted on the Russian Language Affective speech database. The results show that the appropriate use of emotional speech in speaker model training improves the robustness of a speaker recognition system – both when tested on neutral and emotional speech.

Analysis Discrete Hartley Transform for the recognition of female voice based on voice register in singing techniques

Automatic speech recognition is the ability to receive and identify spoken words by converting analog signals to digital and extracting unique vocal characteristics such as pitch, frequency, tone, and rhythm to form speaker models or sound samples. The voice sample used is the voice register, the voice register is the division of the area of the human voice based on the source of the sound, the sensation of the resonant space, shape, color, sound timbre, and the high and low tone produced. Discrete Hartley Transform is used as a transformation to process the sound sample to be classified. DHT, DHT + High Pass Filter and DHT + Low Pass Filter in transforming voice register signals can only classify with an average true positive rate of 69.67%. The addition of the filter does not affect the classification results because the sound frequency used is in ideal conditions so that there is no noise that affects the classification results of the voice register.

Speaker identification using convolutional-long short-term memory neural networks

Speaker identification (SI) techniques has been used in numerous commercial products over the last decades. In SI, the main purpose is to match a voice sample from an unknown speaker to one of the labeled speaker models. To be able accomplish this task, there are two operational phases, training (can be also termed as enrollment) and testing. In both phases, feature extraction and feature matching are the two key steps. In this work, we have extracted features with Mel-Frequency Cepstrum Coefficients (MFCC) because MFCC has accurate representation of the vocal tract, and a Spectral-Flux based Voice Activity Detector (VAD) is implemented to extract features from the speech segments. In feature matching task, we build a Convolutional Long short-term memory (LSTM) Neural Network for the speaker models. We examine main performance of the system in terms of identification rate and compare the proposed method with other SI methods under several noisy conditions at different signal to noise ratio (SNR) levels.Speaker identification (SI) techniques has been used in numerous commercial products over the last decades. In SI, the main purpose is to match a voice sample from an unknown speaker to one of the labeled speaker models. To be able accomplish this task, there are two operational phases, training (can be also termed as enrollment) and testing. In both phases, feature extraction and feature matching are the two key steps. In this work, we have extracted features with Mel-Frequency Cepstrum Coefficients (MFCC) because MFCC has accurate representation of the vocal tract, and a Spectral-Flux based Voice Activity Detector (VAD) is implemented to extract features from the speech segments. In feature matching task, we build a Convolutional Long short-term memory (LSTM) Neural Network for the speaker models. We examine main performance of the system in terms of identification rate and compare the proposed method with other SI methods under several noisy conditions at different signal to noise ratio (SNR) levels.

Fearless steps, NASA’s first heroes: Conversational speech analysis of the Apollo-11 mission control personnel

Between 1963 and 1972, a massive team of dedicated scientists, engineers, and specialists at the NASA Mission Control Center (MCC) worked seamlessly together in a cohesive manner to successfully carry out multiple manned missions to the moon. All communications between personnel were carried out over multiple inter-connected audio channels and recorded on two 30-track analog tapes. Digitization of the entire Apollo-11 mission tapes made possible through the UTDallas-CRSS Fearless Steps (FS) initiative contains the recordings of all the crew members including the three astronauts. With over 600 speakers in constant communication ensuring the astronauts’ safety throughout the mission, the success of the Apollo missions can be attributed largely to the MCC crew members. The focus of this effort is thus, to analyze their high-level group dynamics and intricate communication characteristics and gain insight into the success parameters involved in such large-scale time-critical operations. The 100 hour FS Challenge corpus highlights multiple salient moments like Lift-Off and Lunar-Landing and poses its’ own challenges for core speech tasks. This effort aims at analyzing spontaneous multi-speaker conversations in recordings often degraded by various noise types. Analysis of speech characteristics under varying stress, overlap, and noise conditions are observed to develop novel speech-activity and speaker models. These domain-specific models are leveraged to develop state-of-the-art SLT systems.

Experimental modeling and application of push-pull electrostatic speakers.

This study modeled the circular diaphragm of push-pull electrostatic speakers using electro-mechano-acoustical equivalent circuits to determine the frequency responses related to the average displacement and sound pressure of the diaphragm with respect to acoustic impedance effects. A laser vibrometer system was used to measure the displacement in the center of the circular diaphragms (60 mm and 65 mm in diameter) in optically transparent indium tin oxide (ITO) electrostatic speakers. The sound pressure levels (SPLs) were predicted using the average displacement of the diaphragm without resorting to sound field measurements or finite element analysis. The predicted SPL results were then compared to acoustic measurements. The results demonstrate that ITO plates are a practical tool for the measurement of displacement and the prediction of SPLs. When the proposed electrostatic speaker (65-mm-diameter diaphragm) design was applied to the over-ear electrostatic headphone, the SPL curve obtained from artificial ear measurements was similar to the preferred target curve of Harman International (Stamford, CT). This resulted in outstanding sound quality comparable to that of high-end commercial dynamic headphones.

Speaker tracking using SincNet for multi-channel naturalistic audio corpora: Apollo 11 fearless steps

Apollo-11 was the first manned space mission to successfully bring astronauts to the moon and return them safely. As a massive collaborative effort, with astronauts flying the missions in outer space, the entire communications between flight controllers, their backroom support teams and, astronauts have taken place inside NASA Mission Control Center. The communications stemming from the experienced team of scientists, engineers, and technicians who worked collectively behind the scenes in the MOCR have not been identified. In this study, we propose to identify the speakers in the Mission Control Room for a small subset of 100 h derived from a collective 19 000 h of Apollo-11 audio data, which correspond to three challenging phases of the mission: lift-off, lunar landing, and lunar walking. The speaker identification solution explores a convolutional neural network architecture called SincNet across all three Apollo-11 phases. The speaker models obtained from the subset of 100 hours will be used to track speakers on the complete corpora. Our goal is to make it possible to be able to access each individual speaker involved with the Apollo 11 mission.Apollo-11 was the first manned space mission to successfully bring astronauts to the moon and return them safely. As a massive collaborative effort, with astronauts flying the missions in outer space, the entire communications between flight controllers, their backroom support teams and, astronauts have taken place inside NASA Mission Control Center. The communications stemming from the experienced team of scientists, engineers, and technicians who worked collectively behind the scenes in the MOCR have not been identified. In this study, we propose to identify the speakers in the Mission Control Room for a small subset of 100 h derived from a collective 19 000 h of Apollo-11 audio data, which correspond to three challenging phases of the mission: lift-off, lunar landing, and lunar walking. The speaker identification solution explores a convolutional neural network architecture called SincNet across all three Apollo-11 phases. The speaker models obtained from the subset of 100 hours will be used to track s...

An overview of speaker recognition

Speaker recognition has been studied for many years and has been a hot topic. This paper presents an overview of speaker recognition methods, which include the classical and the state-of-art methods. According to the modular components of speaker recognition system, we fi rstly introduced the fundamentals of speaker recognition, which are mainly divided into two parts: feature extraction and speaker modeling. The most commonly speech features used in speaker recognition were elaborated fi rstly. In particular, the recent progress of deep neural network proposes a new approach of feature extraction and has become the technology trend. Secondly, the classical approaches of speaker recognition model were introduced, and elaborated the recent progress of deep learning speaker recognition. This paper especially provides an in-depth analysis on end-to-end model which consists of a training component to extract features, an enrollment component to training the speaker model, and an evaluation component with appropriate loss function for optimization. The fi nal part concludes the paper with discussion on future trends.

Linguistic Perspectives and Cultural Status of King Gojong in Russian Diplomatic Mission

Linguistic and cultural diversity is inherent in many societies around the world and, despite its importance, this diversity is typically neglected in many educational settings. In the field of language education, the historical prevalence of the monolingual theoretical framework has corroborated with the notion that learners should attain language proficiency based on the native speaker model, which has been mistakenly used as reference for language development. The signing of the peace treaty between Japan and China in 1895 led to the fact that Japan took control of the Korean state and carried out socio-economic reform in this country in its own Japan interests. Russia left unanswered the requests from King of Korea Gojong to the Russian government to send military instructors and advisers. In Korea, Queen Ming, who was brutally murdered by the Japanese in 1895, was a symbol of opposition to the policy pursued by the Japanese state. After these tragic events, King Gojong, together with the heir to the throne, decided to move to the Russian diplomatic mission and continued to remain there until October 1987. Attempts by King Gojong and Queen Ming to maintain the independence of the Korean state, as well as the move of the king along with the heir to the Russian diplomatic mission are assessed differently in South Korean and North Korean historiography now. Overall, due to the multidisciplinary nature of the analytical tools and theoretical frameworks that Cultural Linguistics draws upon, it has significant potential to continue to shed substantial light on the nature of the relationship between language, culture, and conceptualisation.

Digital Forensic Practices and Methodologies for AI Speaker Ecosystems

Various Internet of Things (IoT) devices, such as AI speakers, are being released with different functions to improve user convenience and better life. An AI speaker ecosystem is a cloud-based IoT system built around an AI speaker and IoT devices. In the near future, citizens in whole countries worldwide will be helped in real life when AI-equipped devices are deployed in their homes. Typically, because AI speakers are always operating, they can be used to provide vital evidence for digital forensics; however, privacy issues may arise. AI speakers have provided evidence of murders in the United States and Mexico and are being released without specific regulatory guidelines. In this study, we propose five digital forensic analysis methods for four AI speaker models from different manufacturers released in the Republic of Korea. The five proposed methods are applied to all the AI speaker models, and the results are presented in the Appendix. In particular, we developed a forensic tool for collecting user command history for NAVER Clova.

A framework for speaker retrieval and identification through unsupervised learning

Speaker recognition is a task of remarkable relevance, with applications in diversified domains. Recently, mainly due to the facilities in audio-visual content acquisition, the capacity of analyzing growing datasets independent of labeled data has become a crucial advantage. This paper presents a speaker recognition approach based on recent unsupervised learning methods, which do not require any labeled data or user intervention. The approach is organized in terms of a framework which exploits a rank-based formulation. The similarity information defined by speaker modeling techniques is encoded in ranked lists, which are used as input by the unsupervised learning algorithms. Vector quantization, Gaussian mixture models and i-vectors are employed as modeling techniques, while the algorithms RL-Sim and ReckNN are used for unsupervised learning tasks. The framework was experimentally evaluated on query-by-example speaker retrieval and speaker identification tasks, both on clean and noisy speech recordings. An experimental evaluation was conducted on three public datasets, different languages, and recordings conditions. Effectiveness gains up to +56% on retrieval measures were obtained through the use of unsupervised learning algorithms over traditional speaker recognition techniques.

Characterization of inter-speaker articulatory variability: A two-level multi-speaker modelling approach based on MRI data.

Speech communication relies on articulatory and acoustic codes shared between speakers and listeners despite inter-individual differences in morphology and idiosyncratic articulatory strategies. This study addresses the long-standing problem of characterizing and modelling speaker-independent articulatory strategies and inter-speaker articulatory variability. It explores a multi-speaker modelling approach based on two levels: statistically-based linear articulatory models, which capture the speaker-specific articulatory variability on the one hand, are in turn controlled by a speaker model, which captures the inter-speaker variability on the other hand. A low dimensionality speaker model is obtained by taking advantage of the inter-speaker correlations between morphology and strategy. To validate this approach, contours of the vocal tract articulators were manually segmented on midsagittal MRI data recorded from 11 French speakers uttering 62 vowels and consonants. Using these contours, multi-speaker models with 14 articulatory components and two morphology and strategy components led to overall variance explanations of 66%-69% and root-mean-square errors of 0.36-0.38 cm obtained in leave-one-out procedure over the speakers. Results suggest that inter-speaker variability is more related to the morphology than to the idiosyncratic strategies and illustrate the adaptation of the articulatory components to the morphology.

A Speaker Verification Method Based on TDNN–LSTMP

In speaker recognition, a robust recognition method is essential. This paper proposes a speaker verification method that is based on the time-delay neural network (TDNN) and long short-term memory with recurrent project layer (LSTMP) model for the speaker modeling problem in speaker verification. In this work, we present the application of the fusion of TDNN and LSTMP to the i-vector speaker recognition system that is based on the Gaussian mixture model-universal background model. By using a model that can establish long-term dependencies to create a universal background model that contains a larger amount of speaker information, it is possible to extract more feature parameters, which are speaker dependent, from the speech signal. We conducted experiments with this method on four corpora: two in Chinese and two in English. The equal error rate, minimum detection cost function and detection error tradeoff curve are used as criteria for system performance evaluation. The experimental results show that the TDNN–LSTMP/i-vector speaker recognition method outperforms the baseline system on both Chinese and English corpora and has better robustness.

Exploring the spread of English language learning in South Korea and reflections of the diversifying sociolinguistic context for future English language teaching practices

ABSTRACTThe purpose of this article is to provide an understanding of why traditional English language teaching (ELT) which focuses on native speaker models has become obsolete in South Korea (henceforth Korea). Korea has been known as a linguistically and ethnically homogeneous nation throughout its history. However, the current sociolinguistic situation of Korea tells a different story. Various patterns of migration have emerged around the world and Korea has been no exception. This article explores the widespread passion for English language learning in Korea and the role of English on the individual and nationwide levels. Then, it argues that monolingual and standard language ideologies do not fit into this globalized era. Given the steadily diversifying demographics in Korea and the spread of global English, ELT in Korea ought to reflect English used in the ‘real world’.

Syllables constitute proximate units for Mandarin speakers: Electrophysiological evidence from a masked priming task.

Languages may differ regarding the primary mental unit of phonological encoding in spoken production, with models of speakers of Indo-European languages generally assuming a central role for phonemes, but spoken Chinese production potentially attributing a more prominent role to syllables. In the present study, native Mandarin Chinese speakers named objects that were preceded by briefly presented and masked prime words, which were form related and either matched or mismatched concerning their syllabic structure, or were unrelated. Behavioral results showed a previously reported interaction between prime and target syllable type. Concurrently recorded EEG also exhibited this interaction and further revealed that syllable overlap modulated ERPs mainly in the time window of 300-400ms after picture onset. By contrast, phonemic overlap modulated ERPs from 500ms to 600ms. This pattern might suggest that speakers retrieved syllables before phonemes and strengthens the claim that for Chinese individuals syllables constitute primary functional representations ("proximate units").

Investigating Text-Independent Speaker Verification Systems Under Varied Data Conditions

This work makes an investigation into speaker verification (SV) from the view of practical systems. Limited data SV is preferred in order to have user comfort and effective decision delivery for regular usage. However, reduction in speech data affects the SV performance that becomes a concern for field deployment. In this work, varied data conditions for SV are explored, and sufficient train with limited test data is presented as a preferable anatomy for practical systems. Different explorations are made from the perspective of improving performance in varied data conditions. These explorations include vocal tract constriction feature to include speaker-specific acoustic–phonetic information, different attributes of voice source features that carry alternative/complementary information from that carried by conventional mel-frequency cepstral coefficient features. Further, kernel discriminant analysis is performed at the back end of i-vector-based speaker modeling for channel/session compensation that is found to work well for varied data conditions. Finally, a framework is proposed in combination with the stated explorations to have a better speaker characterization, which is more effective in case of sufficient train and limited test data scenario. The proposed framework achieves significant improvement in performance [equal error rate (EER): 11.20%, detection cost function (DCF): 0.1990], compared to the baseline (EER: 22.31%, DCF: 0.4128) for sufficient train with 2-s test segment case, showing scope toward application-oriented systems.

Speaker Recognition Based on Long-Term Acoustic Features With Analysis Sparse Representation

The performance of a speaker recognition system depends highly on which acoustic features are used. Most speaker recognition systems use short-term acoustic features extracted from a single speech frame, and the most popular short-term acoustic features are the Mel-frequency cepstral coefficients (MFCCs). The short-term features are generally static features no dynamic information in the speech signal is included in either cepstral coefficients or an MFCCs frame. Using an analysis sparse representation model, in this paper, we introduce the long-term acoustic (LTA) feature for text-independent speaker recognition, which is a sparse presentation of the static features and dynamic information for the speaker’s speech. First, the speech signal is segmented into frames which are overlapping with each other, and then the MFCCs frame features can be extracted to construct some super MFCCs frames by stacking some following frames of the current frame to capture the dynamic information of the speech signal. The super MFCCs frames can be combined into a 2-D MFCCs features map (MFCCsmap). Finally, the speaker model can be built based on the analysis sparse model and the sparse representations of the MFCCsmap are used as the LTA features. A state-of-the-art deep neural network (DNN) is employed as a classifier for speaker recognition. The experimental results illustrate the effectiveness and robustness of the proposed system.

Bone- and air-conduction speech combination method for speaker recognition

In this paper, first, we report speaker recognition performance using bone-conduction speech based on an i-vector-based speaker recognition system, which is the current state-of-the-art method. In addition, we propose three speaker recognition methods combining bone-conduction speech and air-conduction speech: a feature combination method, a speaker model combination method, and a similarity score combination method. To evaluate the proposed methods, we conducted speaker recognition experiments using a part of a large speech corpus constructed by the National Research Institute of Police Science, Japan. Experimental results show the bone-conduction speech performs almost the same as the air-conduction speech when the enrolment data and evaluation data are collected in the same session. In addition, all proposed methods improved the speaker recognition performance of air- and bone-conduction speech in the experiments. From these results, we conclude that fusing air- and bone-conduction speech improves the speaker recognition performance.