Two-Terminal Lithium-Mediated Artificial Synapses with Enhanced Weight Modulation for Feasible Hardware Neural Networks

Recently,artificial synapses involving an electrochemical reaction of Li-ion have been attributed to have remarkable synaptic properties.Three-terminal synaptic transistors utilizing Li-ion intercalation exhibits reliable synaptic characteristics by exploiting the advantage of nondistributed weight updates owing to stable ion migrations.However,the three-terminal configurations with large and complex structures impede the crossbar array implementation required for hardware neuromorphic systems.Meanwhile,achieving adequate synaptic performances through effective Li-ion intercalation in vertical two-terminal synaptic devices for array integration remains challenging.Here,two-terminal Au/LixCoO_(2)/Pt artificial synapses are proposed with the potential for practical implementation of hardware neural networks.The Au/LixCoO_(2)/Pt devices demonstrated extraordinary neuromorphic behaviors based on a progressive dearth of Li in LixCoO_(2)films.The intercalation and deintercalation of Li-ion inside the films are precisely controlled over the weight control spike,resulting in improved weight control functionality.Various types of synaptic plasticity were imitated and assessed in terms of key factors such as nonlinearity,symmetricity,and dynamic range.Notably,the LixCoO_(2)-based neuromorphic system outperformed three-terminal synaptic transistors in simulations of convolutional neural networks and multilayer perceptrons due to the high linearity and low programming error.These impressive performances suggest the vertical two-terminal Au/LixCoO_(2)/Pt artificial synapses as promising candidates for hardware neural networks.

Comparing Fine-Tuning, Zero and Few-Shot Strategies with Large Language Models in Hate Speech Detection in English

Large Language Models(LLMs)are increasingly demonstrating their ability to understand natural language and solve complex tasks,especially through text generation.One of the relevant capabilities is contextual learning,which involves the ability to receive instructions in natural language or task demonstrations to generate expected outputs for test instances without the need for additional training or gradient updates.In recent years,the popularity of social networking has provided a medium through which some users can engage in offensive and harmful online behavior.In this study,we investigate the ability of different LLMs,ranging from zero-shot and few-shot learning to fine-tuning.Our experiments show that LLMs can identify sexist and hateful online texts using zero-shot and few-shot approaches through information retrieval.Furthermore,it is found that the encoder-decoder model called Zephyr achieves the best results with the fine-tuning approach,scoring 86.811%on the Explainable Detection of Online Sexism(EDOS)test-set and 57.453%on the Multilingual Detection of Hate Speech Against Immigrants and Women in Twitter(HatEval)test-set.Finally,it is confirmed that the evaluated models perform well in hate text detection,as they beat the best result in the HatEval task leaderboard.The error analysis shows that contextual learning had difficulty distinguishing between types of hate speech and figurative language.However,the fine-tuned approach tends to produce many false positives.

Multi-Objective Equilibrium Optimizer for Feature Selection in High-Dimensional English Speech Emotion Recognition

Speech emotion recognition(SER)uses acoustic analysis to find features for emotion recognition and examines variations in voice that are caused by emotions.The number of features acquired with acoustic analysis is extremely high,so we introduce a hybrid filter-wrapper feature selection algorithm based on an improved equilibrium optimizer for constructing an emotion recognition system.The proposed algorithm implements multi-objective emotion recognition with the minimum number of selected features and maximum accuracy.First,we use the information gain and Fisher Score to sort the features extracted from signals.Then,we employ a multi-objective ranking method to evaluate these features and assign different importance to them.Features with high rankings have a large probability of being selected.Finally,we propose a repair strategy to address the problem of duplicate solutions in multi-objective feature selection,which can improve the diversity of solutions and avoid falling into local traps.Using random forest and K-nearest neighbor classifiers,four English speech emotion datasets are employed to test the proposed algorithm(MBEO)as well as other multi-objective emotion identification techniques.The results illustrate that it performs well in inverted generational distance,hypervolume,Pareto solutions,and execution time,and MBEO is appropriate for high-dimensional English SER.

An Adaptive Hate Speech Detection Approach Using Neutrosophic Neural Networks for Social Media Forensics

Detecting hate speech automatically in social media forensics has emerged as a highly challenging task due tothe complex nature of language used in such platforms. Currently, several methods exist for classifying hatespeech, but they still suffer from ambiguity when differentiating between hateful and offensive content and theyalso lack accuracy. The work suggested in this paper uses a combination of the Whale Optimization Algorithm(WOA) and Particle Swarm Optimization (PSO) to adjust the weights of two Multi-Layer Perceptron (MLPs)for neutrosophic sets classification. During the training process of the MLP, the WOA is employed to exploreand determine the optimal set of weights. The PSO algorithm adjusts the weights to optimize the performanceof the MLP as fine-tuning. Additionally, in this approach, two separate MLP models are employed. One MLPis dedicated to predicting degrees of truth membership, while the other MLP focuses on predicting degrees offalse membership. The difference between these memberships quantifies uncertainty, indicating the degree ofindeterminacy in predictions. The experimental results indicate the superior performance of our model comparedto previous work when evaluated on the Davidson dataset.

Exploring Sequential Feature Selection in Deep Bi-LSTM Models for Speech Emotion Recognition

Machine Learning(ML)algorithms play a pivotal role in Speech Emotion Recognition(SER),although they encounter a formidable obstacle in accurately discerning a speaker’s emotional state.The examination of the emotional states of speakers holds significant importance in a range of real-time applications,including but not limited to virtual reality,human-robot interaction,emergency centers,and human behavior assessment.Accurately identifying emotions in the SER process relies on extracting relevant information from audio inputs.Previous studies on SER have predominantly utilized short-time characteristics such as Mel Frequency Cepstral Coefficients(MFCCs)due to their ability to capture the periodic nature of audio signals effectively.Although these traits may improve their ability to perceive and interpret emotional depictions appropriately,MFCCS has some limitations.So this study aims to tackle the aforementioned issue by systematically picking multiple audio cues,enhancing the classifier model’s efficacy in accurately discerning human emotions.The utilized dataset is taken from the EMO-DB database,preprocessing input speech is done using a 2D Convolution Neural Network(CNN)involves applying convolutional operations to spectrograms as they afford a visual representation of the way the audio signal frequency content changes over time.The next step is the spectrogram data normalization which is crucial for Neural Network(NN)training as it aids in faster convergence.Then the five auditory features MFCCs,Chroma,Mel-Spectrogram,Contrast,and Tonnetz are extracted from the spectrogram sequentially.The attitude of feature selection is to retain only dominant features by excluding the irrelevant ones.In this paper,the Sequential Forward Selection(SFS)and Sequential Backward Selection(SBS)techniques were employed for multiple audio cues features selection.Finally,the feature sets composed from the hybrid feature extraction methods are fed into the deep Bidirectional Long Short Term Memory(Bi-LSTM)network to discern emotions.Since the deep Bi-LSTM can hierarchically learn complex features and increases model capacity by achieving more robust temporal modeling,it is more effective than a shallow Bi-LSTM in capturing the intricate tones of emotional content existent in speech signals.The effectiveness and resilience of the proposed SER model were evaluated by experiments,comparing it to state-of-the-art SER techniques.The results indicated that the model achieved accuracy rates of 90.92%,93%,and 92%over the Ryerson Audio-Visual Database of Emotional Speech and Song(RAVDESS),Berlin Database of Emotional Speech(EMO-DB),and The Interactive Emotional Dyadic Motion Capture(IEMOCAP)datasets,respectively.These findings signify a prominent enhancement in the ability to emotional depictions identification in speech,showcasing the potential of the proposed model in advancing the SER field.

Audio-Text Multimodal Speech Recognition via Dual-Tower Architecture for Mandarin Air Traffic Control Communications

In air traffic control communications (ATCC), misunderstandings between pilots and controllers could result in fatal aviation accidents. Fortunately, advanced automatic speech recognition technology has emerged as a promising means of preventing miscommunications and enhancing aviation safety. However, most existing speech recognition methods merely incorporate external language models on the decoder side, leading to insufficient semantic alignment between speech and text modalities during the encoding phase. Furthermore, it is challenging to model acoustic context dependencies over long distances due to the longer speech sequences than text, especially for the extended ATCC data. To address these issues, we propose a speech-text multimodal dual-tower architecture for speech recognition. It employs cross-modal interactions to achieve close semantic alignment during the encoding stage and strengthen its capabilities in modeling auditory long-distance context dependencies. In addition, a two-stage training strategy is elaborately devised to derive semantics-aware acoustic representations effectively. The first stage focuses on pre-training the speech-text multimodal encoding module to enhance inter-modal semantic alignment and aural long-distance context dependencies. The second stage fine-tunes the entire network to bridge the input modality variation gap between the training and inference phases and boost generalization performance. Extensive experiments demonstrate the effectiveness of the proposed speech-text multimodal speech recognition method on the ATCC and AISHELL-1 datasets. It reduces the character error rate to 6.54% and 8.73%, respectively, and exhibits substantial performance gains of 28.76% and 23.82% compared with the best baseline model. The case studies indicate that the obtained semantics-aware acoustic representations aid in accurately recognizing terms with similar pronunciations but distinctive semantics. The research provides a novel modeling paradigm for semantics-aware speech recognition in air traffic control communications, which could contribute to the advancement of intelligent and efficient aviation safety management.

Analysis on the Translation Methods of the Reported Speech in German Academic Papers-Taking the Translation of“Die Internationalisierung der deutschen Hochschulen”as an Example

Reporting is essential in language use,including the re-expression of other people’s or self’s words,opinions,psychological activities,etc.Grasping the translation methods of reported speech in German academic papers is very important to improve the accuracy of academic paper translation.This study takes the translation of“Internationalization of German Universities”(Die Internationalisierung der deutschen Hochschulen),an academic paper of higher education,as an example to explore the translation methods of reported speech in German academic papers.It is found that the use of word order conversion,part of speech conversion and split translation methods can make the translation more accurate and fluent.This paper helps to grasp the rules and characteristics of the translation of reported speech in German academic papers,and also provides a reference for improving the quality of German-Chinese translation.

Chaotic Elephant Herd Optimization with Machine Learning for Arabic Hate Speech Detection

In recent years,the usage of social networking sites has considerably increased in the Arab world.It has empowered individuals to express their opinions,especially in politics.Furthermore,various organizations that operate in the Arab countries have embraced social media in their day-to-day business activities at different scales.This is attributed to business owners’understanding of social media’s importance for business development.However,the Arabic morphology is too complicated to understand due to the availability of nearly 10,000 roots and more than 900 patterns that act as the basis for verbs and nouns.Hate speech over online social networking sites turns out to be a worldwide issue that reduces the cohesion of civil societies.In this background,the current study develops a Chaotic Elephant Herd Optimization with Machine Learning for Hate Speech Detection(CEHOML-HSD)model in the context of the Arabic language.The presented CEHOML-HSD model majorly concentrates on identifying and categorising the Arabic text into hate speech and normal.To attain this,the CEHOML-HSD model follows different sub-processes as discussed herewith.At the initial stage,the CEHOML-HSD model undergoes data pre-processing with the help of the TF-IDF vectorizer.Secondly,the Support Vector Machine(SVM)model is utilized to detect and classify the hate speech texts made in the Arabic language.Lastly,the CEHO approach is employed for fine-tuning the parameters involved in SVM.This CEHO approach is developed by combining the chaotic functions with the classical EHO algorithm.The design of the CEHO algorithm for parameter tuning shows the novelty of the work.A widespread experimental analysis was executed to validate the enhanced performance of the proposed CEHOML-HSD approach.The comparative study outcomes established the supremacy of the proposed CEHOML-HSD model over other approaches.

Research on the Application of Second Language Acquisition Theory in College English Speech Teaching

The teaching of English speeches in universities aims to enhance oral communication ability,improve English communication skills,and expand English knowledge,occupying a core position in English teaching in universities.This article takes the theory of second language acquisition as the background,analyzes the important role and value of this theory in English speech teaching in universities,and explores how to apply the theory of second language acquisition in English speech teaching in universities.It aims to strengthen the cultivation of English skilled talents and provide a brief reference for improving English speech teaching in universities.

Capacitated stochastic coloured Petri net-based approach for computing two-terminal reliability of multi-state network

Classical network reliability problems assume both net- works and components have only binary states, fully working or fully failed states. But many actual networks are multi-state, such as communication networks and transportation networks. The nodes and arcs in the networks may be in intermediate states which are not fully working either fully failed. A simulation ap- proach for computing the two-terminal reliability of a multi-state network is described. Two-terminal reliability is defined as the probability that d units of demand can be supplied from the source to sink nodes under the time threshold T. The capacities of arcs may be in a stochastic state following any discrete or continuous distribution. The transmission time of each arc is also not a fixed number but stochastic according to its current capacity and de- mand. To solve this problem, a capacitated stochastic coloured Petri net is proposed for modelling the system behaviour. Places and transitions respectively stand for the nodes and arcs of a net- work. Capacitated transition and self-modified token colour with route information are defined to describe the multi-state network. By the simulation, the two-terminal reliability and node importance can be estimated and the optimal route whose reliability is highest can also be given. Finally, two examples of different kinds of multi- state networks are given.

Design of Hierarchical Classifier to Improve Speech Emotion Recognition

Automatic Speech Emotion Recognition(SER)is used to recognize emotion from speech automatically.Speech Emotion recognition is working well in a laboratory environment but real-time emotion recognition has been influenced by the variations in gender,age,the cultural and acoustical background of the speaker.The acoustical resemblance between emotional expressions further increases the complexity of recognition.Many recent research works are concentrated to address these effects individually.Instead of addressing every influencing attribute individually,we would like to design a system,which reduces the effect that arises on any factor.We propose a two-level Hierarchical classifier named Interpreter of responses(IR).Thefirst level of IR has been realized using Support Vector Machine(SVM)and Gaussian Mixer Model(GMM)classifiers.In the second level of IR,a discriminative SVM classifier has been trained and tested with meta information offirst-level classifiers along with the input acoustical feature vector which is used in primary classifiers.To train the system with a corpus of versatile nature,an integrated emotion corpus has been composed using emotion samples of 5 speech corpora,namely;EMO-DB,IITKGP-SESC,SAVEE Corpus,Spanish emotion corpus,CMU's Woogle corpus.The hierarchical classifier has been trained and tested using MFCC and Low-Level Descriptors(LLD).The empirical analysis shows that the proposed classifier outperforms the traditional classifiers.The proposed ensemble design is very generic and can be adapted even when the number and nature of features change.Thefirst-level classifiers GMM or SVM may be replaced with any other learning algorithm.

Multilayer Neural Network Based Speech Emotion Recognition for Smart Assistance

Day by day,biometric-based systems play a vital role in our daily lives.This paper proposed an intelligent assistant intended to identify emotions via voice message.A biometric system has been developed to detect human emotions based on voice recognition and control a few electronic peripherals for alert actions.This proposed smart assistant aims to provide a support to the people through buzzer and light emitting diodes(LED)alert signals and it also keep track of the places like households,hospitals and remote areas,etc.The proposed approach is able to detect seven emotions:worry,surprise,neutral,sadness,happiness,hate and love.The key elements for the implementation of speech emotion recognition are voice processing,and once the emotion is recognized,the machine interface automatically detects the actions by buzzer and LED.The proposed system is trained and tested on various benchmark datasets,i.e.,Ryerson Audio-Visual Database of Emotional Speech and Song(RAVDESS)database,Acoustic-Phonetic Continuous Speech Corpus(TIMIT)database,Emotional Speech database(Emo-DB)database and evaluated based on various parameters,i.e.,accuracy,error rate,and time.While comparing with existing technologies,the proposed algorithm gave a better error rate and less time.Error rate and time is decreased by 19.79%,5.13 s.for the RAVDEES dataset,15.77%,0.01 s for the Emo-DB dataset and 14.88%,3.62 for the TIMIT database.The proposed model shows better accuracy of 81.02%for the RAVDEES dataset,84.23%for the TIMIT dataset and 85.12%for the Emo-DB dataset compared to Gaussian Mixture Modeling(GMM)and Support Vector Machine(SVM)Model.

A Multi-Level Circulant Cross-Modal Transformer for Multimodal Speech Emotion Recognition

Speech emotion recognition,as an important component of humancomputer interaction technology,has received increasing attention.Recent studies have treated emotion recognition of speech signals as a multimodal task,due to its inclusion of the semantic features of two different modalities,i.e.,audio and text.However,existing methods often fail in effectively represent features and capture correlations.This paper presents a multi-level circulant cross-modal Transformer(MLCCT)formultimodal speech emotion recognition.The proposed model can be divided into three steps,feature extraction,interaction and fusion.Self-supervised embedding models are introduced for feature extraction,which give a more powerful representation of the original data than those using spectrograms or audio features such as Mel-frequency cepstral coefficients(MFCCs)and low-level descriptors(LLDs).In particular,MLCCT contains two types of feature interaction processes,where a bidirectional Long Short-term Memory(Bi-LSTM)with circulant interaction mechanism is proposed for low-level features,while a two-stream residual cross-modal Transformer block is appliedwhen high-level features are involved.Finally,we choose self-attention blocks for fusion and a fully connected layer to make predictions.To evaluate the performance of our proposed model,comprehensive experiments are conducted on three widely used benchmark datasets including IEMOCAP,MELD and CMU-MOSEI.The competitive results verify the effectiveness of our approach.

Age-related hearing loss accelerates the decline in fast speech comprehension and the decompensation of cortical network connections

Patients with age-related hearing loss face hearing difficulties in daily life.The causes of age-related hearing loss are complex and include changes in peripheral hearing,central processing,and cognitive-related abilities.Furthermore,the factors by which aging relates to hearing loss via changes in audito ry processing ability are still unclear.In this cross-sectional study,we evaluated 27 older adults(over 60 years old) with age-related hearing loss,21 older adults(over 60years old) with normal hearing,and 30 younger subjects(18-30 years old) with normal hearing.We used the outcome of the uppe r-threshold test,including the time-compressed thres h old and the speech recognition threshold in noisy conditions,as a behavioral indicator of auditory processing ability.We also used electroencephalogra p hy to identify presbycusis-related abnormalities in the brain while the participants were in a spontaneous resting state.The timecompressed threshold and speech recognition threshold data indicated significant diffe rences among the groups.In patients with age-related hearing loss,information masking(babble noise) had a greater effect than energy masking(speech-shaped noise) on processing difficulties.In terms of resting-state electroencephalography signals,we observed enhanced fro ntal lobe(Brodmann’s area,BA11) activation in the older adults with normal hearing compared with the younger participants with normal hearing,and greater activation in the parietal(BA7) and occipital(BA19) lobes in the individuals with age-related hearing loss compared with the younger adults.Our functional connection analysis suggested that compared with younger people,the older adults with normal hearing exhibited enhanced connections among networks,including the default mode network,sensorimotor network,cingulo-opercular network,occipital network,and frontoparietal network.These results suggest that both normal aging and the development of age-related hearing loss have a negative effect on advanced audito ry processing capabilities and that hearing loss accele rates the decline in speech comprehension,especially in speech competition situations.Older adults with normal hearing may have increased compensatory attentional resource recruitment represented by the to p-down active listening mechanism,while those with age-related hearing loss exhibit decompensation of network connections involving multisensory integration.

On the Absence of Carrier Drift in Two-Terminal Devices and the Origin of Their Lowest Resistance Per Carrier R_k=h/Q²

After a criticism on today’s model for electrical noise in resistors, we pass to use a Quantum-compliant model based on the discreteness of electrical charge in a complex Admittance. From this new model we show that carrier drift viewed as charged particle motion in response to an electric field is unlike to occur in bulk regions of Solid-State devices where carriers react as dipoles against this field. The absence of the shot noise that charges drifting in resistors should produce and the evolution of the Phase Noise with the active power existing in the resonators of L-C oscillators, are two effects added in proof for this conduction model without carrier drift where the resistance of any two-terminal device becomes discrete and has a minimum value per carrier that is the Quantum Hall resistance Rk=h/q2

Speech Recognition via CTC-CNN Model

In the speech recognition system,the acoustic model is an important underlying model,and its accuracy directly affects the performance of the entire system.This paper introduces the construction and training process of the acoustic model in detail and studies the Connectionist temporal classification(CTC)algorithm,which plays an important role in the end-to-end framework,established a convolutional neural network(CNN)combined with an acoustic model of Connectionist temporal classification to improve the accuracy of speech recognition.This study uses a sound sensor,ReSpeakerMic Array v2.0.1,to convert the collected speech signals into text or corresponding speech signals to improve communication and reduce noise and hardware interference.The baseline acousticmodel in this study faces challenges such as long training time,high error rate,and a certain degree of overfitting.The model is trained through continuous design and improvement of the relevant parameters of the acousticmodel,and finally the performance is selected according to the evaluation index.Excellentmodel,which reduces the error rate to about 18%,thus improving the accuracy rate.Finally,comparative verificationwas carried out from the selection of acoustic feature parameters,the selection of modeling units,and the speaker’s speech rate,which further verified the excellent performance of the CTCCNN_5+BN+Residual model structure.In terms of experiments,to train and verify the CTC-CNN baseline acoustic model,this study uses THCHS-30 and ST-CMDS speech data sets as training data sets,and after 54 epochs of training,the word error rate of the acoustic model training set is 31%,the word error rate of the test set is stable at about 43%.This experiment also considers the surrounding environmental noise.Under the noise level of 80∼90 dB,the accuracy rate is 88.18%,which is the worst performance among all levels.In contrast,at 40–60 dB,the accuracy was as high as 97.33%due to less noise pollution.

Investigation of hearing aid users'speech understanding in noise and their spectral-temporal resolution skills

Purpose:Our study aims to compare speech understanding in noise and spectral-temporal resolution skills with regard to the degree of hearing loss,age,hearing aid use experience and gender of hearing aid users.Methods:Our study included sixty-eight hearing aid users aged between 40-70 years,with bilateral mild and moderate symmetrical sensorineural hearing loss.Random gap detection test,Turkish matrix test and spectral-temporally modulated ripple test were implemented on the participants with bilateral hearing aids.The test results acquired were compared statistically according to different variables and the correlations were examined.Results:No statistically significant differences were observed for speech-in-noise recognition,spectraltemporal resolution among older and younger adults in hearing aid users(p>0.05).There wasn’t found a statistically significant difference among test outcomes as regards different hearing loss degrees(p>0.05).Higher performances were obtained in terms of temporal resolution in male participants and participants with more hearing aid use experience(p<0.05).Significant correlations were obtained between the results of speech-in-noise recognition,temporal resolution and spectral resolution tests performed with hearing aids(p<0.05).Conclusion:Our study findings emphasized the importance of regular hearing aid use and it showed that some auditory skills can be improved with hearing aids.Observation of correlations among the speechin-noise recognition,temporal resolution and spectral resolution tests have revealed that these skills should be evaluated as a whole to maximize the patient’s communication abilities.

Improving Speech Enhancement Framework via Deep Learning

Speech plays an extremely important role in social activities.Many individuals suffer from a“speech barrier,”which limits their communication with others.In this study,an improved speech recognitionmethod is proposed that addresses the needs of speech-impaired and deaf individuals.A basic improved connectionist temporal classification convolutional neural network(CTC-CNN)architecture acoustic model was constructed by combining a speech database with a deep neural network.Acoustic sensors were used to convert the collected voice signals into text or corresponding voice signals to improve communication.The method can be extended to modern artificial intelligence techniques,with multiple applications such as meeting minutes,medical reports,and verbatim records for cars,sales,etc.For experiments,a modified CTC-CNN was used to train an acoustic model,which showed better performance than the earlier common algorithms.Thus a CTC-CNN baseline acoustic model was constructed and optimized,which reduced the error rate to about 18%and improved the accuracy rate.

Emotional Vietnamese Speech Synthesis Using Style-Transfer Learning

In recent years,speech synthesis systems have allowed for the produc-tion of very high-quality voices.Therefore,research in this domain is now turning to the problem of integrating emotions into speech.However,the method of con-structing a speech synthesizer for each emotion has some limitations.First,this method often requires an emotional-speech data set with many sentences.Such data sets are very time-intensive and labor-intensive to complete.Second,training each of these models requires computers with large computational capabilities and a lot of effort and time for model tuning.In addition,each model for each emotion failed to take advantage of data sets of other emotions.In this paper,we propose a new method to synthesize emotional speech in which the latent expressions of emotions are learned from a small data set of professional actors through a Flow-tron model.In addition,we provide a new method to build a speech corpus that is scalable and whose quality is easy to control.Next,to produce a high-quality speech synthesis model,we used this data set to train the Tacotron 2 model.We used it as a pre-trained model to train the Flowtron model.We applied this method to synthesize Vietnamese speech with sadness and happiness.Mean opi-nion score(MOS)assessment results show that MOS is 3.61 for sadness and 3.95 for happiness.In conclusion,the proposed method proves to be more effec-tive for a high degree of automation and fast emotional sentence generation,using a small emotional-speech data set.

A New Speech Encoder Based on Dynamic Framing Approach

Latent information is difficult to get from the text in speech synthesis.Studies show that features from speech can get more information to help text encoding.In the field of speech encoding,a lot of work has been conducted on two aspects.The first aspect is to encode speech frame by frame.The second aspect is to encode the whole speech to a vector.But the scale in these aspects is fixed.So,encoding speech with an adjustable scale for more latent information is worthy of investigation.But current alignment approaches only support frame-by-frame encoding and speech-to-vector encoding.It remains a challenge to propose a new alignment approach to support adjustable scale speech encoding.This paper presents the dynamic speech encoder with a new alignment approach in conjunction with frame-by-frame encoding and speech-to-vector encoding.The speech feature fromourmodel achieves three functions.First,the speech feature can reconstruct the origin speech while the length of the speech feature is equal to the text length.Second,our model can get text embedding fromspeech,and the encoded speech feature is similar to the text embedding result.Finally,it can transfer the style of synthesis speech and make it more similar to the given reference speech.