Artificial Speech Research Articles

Искусственно сгенерированный академический текст (лингвопрагматический аспект)

In the process of integrating ‘artificial intelligence’ (AI) technologies into academic discourse, the procedure of writing an academic text is changing, and a new type of communicative interaction is formed – AI as the author of a ‘generated text’ (GТ) and a human being as the recipient of a text from AI. The aim of the study is to substantiate the perception of such a text by the recipient from the point of view of pragmalinguistic analysis on the basis of the identified features and indicators. An analysis of the functioning of artificially generated academic text is presented (5 scientific texts were generated by the Mahtext system). For the first time machine text is considered as an ‘speech act’ in three directions of its realisation: plan of expression, content and prediction. It is revealed that the neural network as a text author cannot transform and convey any attitude to the addressee (recipient), just as it cannot consciously influence him through its texts and have pragmatic intentions. The novelty of the study consists in the following: 1) for the first time we consider not an anthropological text, but an AI text; 2) as a result of pragmalinguistic analysis are revealed 7 main features and 7 indicators of GT; 3) the concept of ‘artificial speech act’ is proposed for the first time.

Evoking artificial speech perception through invasive brain stimulation for brain-computer interfaces: current challenges and future perspectives.

Encoding artificial perceptions through brain stimulation, especially that of higher cognitive functions such as speech perception, is one of the most formidable challenges in brain-computer interfaces (BCI). Brain stimulation has been used for functional mapping in clinical practices for the last 70 years to treat various disorders affecting the nervous system, including epilepsy, Parkinson's disease, essential tremors, and dystonia. Recently, direct electrical stimulation has been used to evoke various forms of perception in humans, ranging from sensorimotor, auditory, and visual to speech cognition. Successfully evoking and fine-tuning artificial perceptions could revolutionize communication for individuals with speech disorders and significantly enhance the capabilities of brain-computer interface technologies. However, despite the extensive literature on encoding various perceptions and the rising popularity of speech BCIs, inducing artificial speech perception is still largely unexplored, and its potential has yet to be determined. In this paper, we examine the various stimulation techniques used to evoke complex percepts and the target brain areas for the input of speech-like information. Finally, we discuss strategies to address the challenges of speech encoding and discuss the prospects of these approaches.

Design of smart home environment based on wireless sensor system and artificial speech recognition

In the context of the rapid development of intelligent technology, smart home has become the focus of attention. However, the traditional smart home system has some problems such as complicated operation and inconvenient information interaction. In order to solve these problems and improve the user experience of smart home environment, this study establishes a fast and efficient smart home environment design scheme by introducing wireless sensor system and artificial speech recognition technology. By arranging multiple wireless sensor nodes, the monitoring and data acquisition of environmental parameters are realized, and then the voice control and command interaction of smart home devices are realized by using artificial voice recognition technology. Through the effective combination of wireless sensor system and artificial voice recognition technology, users can realize the control of smart home environment through voice commands, and improve the convenience and intelligence of the system. The experimental results show that the design scheme can accurately and stably realize the recognition of user voice commands and the control of smart home devices, and users' evaluation of the degree of intelligence and user experience has also been significantly improved.

A Comparative Evaluation on Data Transformation Approach for Artificial Speech Detection

The rise of voice biometrics has transformed user authentication and offered enhanced security and convenience while phasing out less secure methods. Despite these advancements, Automatic Speaker Verification (ASV) systems remain vulnerable to spoofing, particularly with artificial speech generated swiftly using advanced speech synthesis and voice conversion algorithms. A recent data transformation technique achieved an impressive Equal Error Rate (EER) of 1.42% on the ASVspoof 2019 Logical Access Dataset. While this approach predominantly relies on Support Vector Machine (SVM) as the backend classifier for artificial speech detection, it is vital to explore a broader range of classifiers to enhance resilience. This paper addresses this research gap by systematically assessing classifier efficacy in artificial speech detection. The objectives are twofold: first, to evaluate various classifiers, not limited to SVM, and identify those best suited for artificial speech detection; second, to compare this approach's performance with existing methods. The evaluation demonstrated SVM-Polynomial as the top-performing classifier, surpassing the end-to-end learning approach. This work contributes to a deeper understanding of classifier efficacy and equips researchers and practitioners with a diversified toolkit for building robust ASV spoofing detection systems.

The Application of Intelligent Speech Recognition in the Teaching of Spoken English in Colleges and Universities

Abstract With the rapid development of science and technology, the use of artificial intelligence in the field of higher education is becoming more and more common. The use of intelligent speech recognition technology in English teaching in colleges and universities is on the rise. The research utilizes speech recognition logic to create a hybrid CTC/Attention model for teaching spoken English in colleges and universities. Then we compare and analyze the performance of artificial speech recognition models, such as the HMM model and CTC/Attention model, and explore the effects of teaching English speaking in colleges and universities using this model. In terms of the overall recognition rate, with the increase of the test vocabulary, the recognition rates of both the artificial speech recognition model and the HMM model decrease, while the CTC/Attention model shows an increasing trend. Except for intonation, the CTC/Attention model has the best recognition rate performance across all dimensions. Through the college-speaking teaching experiment based on the CTC/Attention model, the students’ total English speaking performance, accuracy, fluency, and complexity were improved. The Sig values of total speaking achievement, accuracy, fluency, and complexity are less than 0.05. College English-speaking teaching with intelligent speech recognition has a significant positive impact on the improvement of student’s English-speaking achievement, as indicated.

Continuous synthesis of artificial speech sounds from human cortical surface recordings during silent speech production

Objective. Brain–computer interfaces can restore various forms of communication in paralyzed patients who have lost their ability to articulate intelligible speech. This study aimed to demonstrate the feasibility of closed-loop synthesis of artificial speech sounds from human cortical surface recordings during silent speech production. Approach. Ten participants with intractable epilepsy were temporarily implanted with intracranial electrode arrays over cortical surfaces. A decoding model that predicted audible outputs directly from patient-specific neural feature inputs was trained during overt word reading and immediately tested with overt, mimed and imagined word reading. Predicted outputs were later assessed objectively against corresponding voice recordings and subjectively through human perceptual judgments. Main results. Artificial speech sounds were successfully synthesized during overt and mimed utterances by two participants with some coverage of the precentral gyrus. About a third of these sounds were correctly identified by naïve listeners in two-alternative forced-choice tasks. A similar outcome could not be achieved during imagined utterances by any of the participants. However, neural feature contribution analyses suggested the presence of exploitable activation patterns during imagined speech in the postcentral gyrus and the superior temporal gyrus. In future work, a more comprehensive coverage of cortical surfaces, including posterior parts of the middle frontal gyrus and the inferior frontal gyrus, could improve synthesis performance during imagined speech. Significance. As the field of speech neuroprostheses is rapidly moving toward clinical trials, this study addressed important considerations about task instructions and brain coverage when conducting research on silent speech with non-target participants.

The Role of Average Fundamental Frequency Difference on the Intelligibility of Real-Life Competing Sentences.

The average fundamental frequency separation (∆fo) between two competing voices has been shown to provide an important cue for target-speech intelligibility. However, some of the previous investigations used speech materials with linguistic properties and fo characteristics that may not be typical of realistic acoustic scenarios. This study investigated to what extent the effect of ∆fo generalizes to more real-life speech. Real-life sentences and a well-controlled method for manipulating the acoustic stimuli were employed. Fifteen young normal-hearing native Danish listeners were tested in a two-competing-voices sentence recognition task at several target-to-masker ratios (TMRs) and ∆fos. Compared to previous studies that addressed the same experimental scenario with less realistic speech materials, the present results showed only a moderate effect of ∆fo at negative TMRs and a negligible effect at positive TMRs. An analysis of the employed stimuli showed that a large ∆fo effect on the target speech intelligibility is only observed when the competing sentences have highly synchronous fo trajectories, which is typical of the artificial speech materials employed in previous studies. Overall, the present results suggest a relatively small effect of ∆fo on the intelligibility of real-life speech, as compared to previously employed artificial speech, in two-competing-sentences conditions.

Performance Evaluation of Environmental Sound Classification: A Machine Learning Stacking and Multi-Criteria Metrics Based Approach

This study proposes an Environment Sound Classification Task (ESC) model that includes numerous element channels given as a contribution to Machine learning with an Attention instrument. ESC is a significant testing issue. The interest in the paper lies in utilizing different part channels involving the MFCCs-Mel Frequency Cepstral Coefficients a mutual module in speaker detection and artificial speech systems. LPCs-Linear Prediction Coefficients and Linear Prediction Cepstral Coefficients were the most commonly used types in ASR- Automated speech recognition. The paper also discusses some basic features of MFCCs and how to put them into practice. The techniques used for this project are Background Gaussian Noise and Time Shifting, by observing that every technique is implemented with a provided probability, however, when at the time of generation of a new sample the same spectrogram input may have several combinations. We go through the blend information expansion method to additional lift execution. Our model can accomplish cutting-edge execution on three benchmark climate sound characterization datasets, for example, the UrbanSound8K.

End-to-End Video-to-Speech Synthesis Using Generative Adversarial Networks.

Video-to-speech is the process of reconstructing the audio speech from a video of a spoken utterance. Previous approaches to this task have relied on a two-step process where an intermediate representation is inferred from the video and is then decoded into waveform audio using a vocoder or a waveform reconstruction algorithm. In this work, we propose a new end-to-end video-to-speech model based on generative adversarial networks (GANs) which translates spoken video to waveform end-to-end without using any intermediate representation or separate waveform synthesis algorithm. Our model consists of an encoder-decoder architecture that receives raw video as input and generates speech, which is then fed to a waveform critic and a power critic. The use of an adversarial loss based on these two critics enables the direct synthesis of the raw audio waveform and ensures its realism. In addition, the use of our three comparative losses helps establish direct correspondence between the generated audio and the input video. We show that this model is able to reconstruct speech with remarkable realism for constrained datasets such as GRID, and that it is the first end-to-end model to produce intelligible speech for Lip Reading in the Wild (LRW), featuring hundreds of speakers recorded entirely "in the wild." We evaluate the generated samples in two different scenarios-seen and unseen speakers-using four objective metrics which measure the quality and intelligibility of artificial speech. We demonstrate that the proposed approach outperforms all previous works in most metrics on GRID and LRW.

Learning and controlling the source-filter representation of speech with a variational autoencoder

Understanding and controlling latent representations in deep generative models is a challenging yet important problem for analyzing, transforming and generating various types of data. In speech processing, inspiring from the anatomical mechanisms of phonation, the source-filter model considers that speech signals are produced from a few independent and physically meaningful continuous latent factors, among which the fundamental frequency f0 and the formants are of primary importance. In this work, we start from a variational autoencoder (VAE) trained in an unsupervised manner on a large dataset of unlabeled natural speech signals, and we show that the source-filter model of speech production naturally arises as orthogonal subspaces of the VAE latent space. Using only a few seconds of labeled speech signals generated with an artificial speech synthesizer, we propose a method to identify the latent subspaces encoding f0 and the first three formant frequencies, we show that these subspaces are orthogonal, and based on this orthogonality, we develop a method to accurately and independently control the source-filter speech factors within the latent subspaces. Without requiring additional information such as text or human-labeled data, this results in a deep generative model of speech spectrograms that is conditioned on f0 and the formant frequencies, and which is applied to the transformation speech signals. Finally, we also propose a robust f0 estimation method that exploits the projection of a speech signal onto the learned latent subspace associated with f0.

A comparison of different support vector machine kernels for artificial speech detection

As the emergence of the voice biometric provides enhanced security and convenience, voice biometric-based applications such as speaker verification were gradually replacing the authentication techniques that were less secure. However, the automatic speaker verification (ASV) systems were exposed to spoofing attacks, especially artificial speech attacks that can be generated with a large amount in a short period of time using state-of-the-art speech synthesis and voice conversion algorithms. Despite the extensively used support vector machine (SVM) in recent works, there were none of the studies shown to investigate the performance of different SVM settings against artificial speech detection. In this paper, the performance of different SVM settings in artificial speech detection will be investigated. The objective is to identify the appropriate SVM kernels for artificial speech detection. An experiment was conducted to find the appropriate combination of the proposed features and SVM kernels. Experimental results showed that the polynomial kernel was able to detect artificial speech effectively, with an equal error rate (EER) of 1.42% when applied to the presented handcrafted features.

Шляхи вдосконалення комунікативно-орієнтованого усного мовлення студентів англійською мовою (за автентичними матеріалами)

The presented article briefly examines the problems that first and second course students confront while speaking. The article emphasises that English has become a fashionable language to learn in Azerbaijan in the past decades. First course students know grammar rules well enough and they possess rich vocabulary also. Yet most students (not all of them, of course) are unable to take part in simple conversations either with a teacher or with each other. The lack of self-confidence in students lead them to express their view points in Azerbaijani. The presented paper emphasizes the importance of the use of authentic materials to improve first and second course students’ communication-oriented skills in English. It is indicated in the paper that according to the requirements of the communicative approach, the usage of authentic texts and language materials in the English language lesson are better than the presentation of artificial texts. Speaking is the productive use of language and improving students’ speaking skills which is considered to be the most difficult part of language teaching and is directly connected with motivation, communicative exercises, listening activities. In the article, it is stated that habit forming provides an excellent incentive so speak English fluently. With regard to the selection of authentic materials, the interests of the students should be taken into consideration. Those authentic materials should be presented to the students that meet their needs and requirements. In addition, the article outlines the types of authentic materials. Teachers are encouraged to use all the types authentic materials in class. Authentic reading and listening materials used in English classes and the activities to be carried out on them are also grouped. The paper points out that authentic materials have become an important element of the English language lessons noting that students should use them in English more often in the contexts in which real oral speech occurs naturally than in the context in which artificial speech occurs. Therefore, it is important to use authentic materials in the process of teaching English

К ВОПРОСУ О ДОСТОВЕРНОСТИ ИСТОЧНИКА ЖАНРА

The question of the inconsistency of the definition of linguistic reliability and the principles of its identification is debatable. The problem of the authenticity of the genre source has not been raised in the scientific review, but it turns out to be relevant due to the contradictions that arise at the level of choosing a genre description source. The decision of linguists to use any of the available statements to describe the genre is a priori debatable. The contradictions of this approach lie in the difference in the types of information contained in the varieties of speech. The question of the qualification of the type of information lies in the differentiation of natural and artificial speech variants, where the natural variety of speech is characterized by the greatest degree of reliability.

Design of Political Online Teaching Based on Artificial Speech Recognition and Deep Learning.

With the emergence of the information age, computers have entered the homes of ordinary people and have become essential daily appliances for people. The integration of people and computers has become more popular and in-depth. Based on this situation, how to make computers and humans communicate intelligently and make human-computer interaction more convenient and practical is the main issue of scientific research and discussion on computers. Language recognition is widely used in industry, the economy, commerce, tourism, style, and sports. The BP neural network based on deep learning has supercomputing functions and is very suitable for speech recognition, which promotes the promotion and application of speech recognition technology in many fields. Different types of applications and diverse scenarios enable users to experience different personal experiences. This paper examines the use of application software for ideological and political theory courses in colleges and universities and the application of application software for ideological and political theory courses in colleges and universities in the management of ideological and political theory courses. Analyzed the use of apps in the management of ideological and political theory courses in colleges and universities, analyzed the possibilities and challenges that this brought to the management of ideological and political theory courses in colleges and universities based on the app, and put forward the necessary countermeasures. Including the development and design of ideological and political theory applications, combining deep learning and CTC algorithms to build acoustic models, using server-client interaction, designing an ideological and political theory course app based on speech recognition and deep learning, and forming an offline speech recognition system software platform, the app provides a certain reference for improving the skills of teaching managers.

An Approach for Formant Synthesis of Kannada

Formant Synthesis is one of the ways of producing artificial speech using a digital computer. Artificial or synthetic speech is commonly used in Text To Speech (TTS) conversion systems and in various Human Machine Interfaces (HMI). A very good example of usage of synthetic speech is in automatic public address or announcement systems that are commonly found in railway stations, airports, bus stations, banks, malls etc., This paper presents a method of synthesizing Kannada speech utterances such as Vowels, Semi-vowels, Diphthongs, Consonants and Consonant-Vowel (CV) co-articulations that forms the building block of a fully functional Formant Synthesizer for Kannada language. A Klatt type formant synthesizer implemented using C# .NET is employed for producing these utterances in this work. PRAAT tool is used to perform analysis and obtaining the control parameters needed for generating various utterances. A total of 192 utterances including 10 vowels, 2 diphthongs and 180 CV co-articulations were synthesized and of which 176 were correctly identified in perception test giving an accuracy of 91.67% on an average. Formant synthesizer allows more prosody control of synthesized utterances over other methods of speech synthesis and hence this work can serve as a guide to researchers trying to develop a functional format synthesizer for Kannada language.

Measuring the Quality of Low-Resourced Statistical Parametric Speech Synthesis Trained with Noise-Degraded Data Supported by the University of Costa Rica

After the successful implementation of speech synthesis in several languages, the study of robustness became an important topic so as to increase the possibility of building voices from non-standard sources, e.g. historical recordings, children's speech, and data freely available on the Internet. In this work, a measure of the influence of noise in the source speech of the statistical parametric speech synthesis system based on HMM is performed, for a case of a low-resourced database. For this purpose, three types of additive noise were considered at five signal-to-noise ratio levels to affect the source speech data. Using objective measures to assess the perceptual quality of the results and the propagation of the noise through all the processes of building speech synthesis, the results show a severe drop in the quality of artificial speech, even for the cases of lower levels of noise. Such degradation seems to be independent of the noise type, and is at lower proportion to the noise level. This results are of importance for any practical implementation of speech synthesis from degraded data in similar conditions, and shows that applying denoising processes became mandatory in order to keep the possibility of building intelligible voices.

Speech Impaired by Half Masks Used for the Respiratory Tract Protection.

Filtering half masks belong to the group of personal protective equipment in the work environment. They protect the respiratory tract but may hinder breath and suppress speech. The present work is focused on the attenuation of sound by the half masks known as “filtering facepieces”, FFPs, of various construction and filtration efficiency. Rather than study the perception of speech by humans, we used a generator of white noise and artificial speech to obtain objective characteristics of the attenuation. The generator speaker was either covered by an FFP or remained uncovered while a class 1 meter measured sound pressure levels in 1/3 octave bands with center frequencies 100–20 kHz at distances from 1 to 5 m from the speaker. All five FFPs suppressed acoustic waves from the octave bands with center frequencies of 1 kHz and higher, i.e., in the frequency range responsible for 80% of the perceived speech intelligibility, particularly in the 2 kHz-octave band. FFPs of higher filtration efficiency stronger attenuated the sound. Moreover, the FFPs changed the voice timbre because the attenuation depended on the wave frequency. The two combined factors can impede speech intelligibility.

Online Speech Therapy System

Abstract: This paper provides information about online speech therapy system. The power of internet is endless. In today’s world with the advancement of knowledge and technology we can achieve any challenge in life. This paper is about providing assistance to people with speech disorders. Speech disorder is a common problem world wide and some people with this disorder face many challenges and embarrassment in their life. This paper aims at providing assistance to these people with artificial intelligence technology used. Keywords: Artificial intelligence, neural network, speech and language disorder, speech and language assistance

Acoustical testing of face coverings using an artificial acoustic head with speech audio

The COVID-19 pandemic has caused many communities to require or at least recommend the wearing of face coverings, masks, and/or shields inside indoor public spaces to reduce the transmission rate of the virus. Several types of face coverings are studied, including standardized N95 and KN95, assorted cloth, and clear plastic face shields. They are evaluated on an acoustical head and torso simulator (HATS) setup in a classroom with two different locations for monitoring via a sound level meter. The HATS is used as a controlled and repeatable artificial voice or sound source, which reproduces the ITU-T artificial speech signals as well as playback of real speech signals for American English, male and female voices. The signals are recorded with a sound level meter at a distance of 2 and 6 m between source and receiver in a classroom environment. These signals evaluated in time and frequency domains to better understand the acoustical difference between the no-mask and various masks conditions. These are compared with the same classroom environment but using white noise test signals.

Performance analysis of different intonation models in Kannada speech synthesis

Text <span lang="EN-US">to speech (TTS) is a system that generates artificial speech from text input. The prosodic models used improve the quality of the synthesized speech especially naturalness and intelligibility. The prosody involves intonation, intonation refers to the variations in the pitch frequency (F0) with respect to time in an utterance. This work mainly concentrates on building feedback neural network model to predict F0 contour in the utterances using Fujisaki intonation model parameters as the input features to the network since the Fujisaki intonation model is data driven and not a rule based one. In this work we have built 4-layer feedback neural network in the festival framework. Finally, the synthetically generated Kannada speech using the neural network model, is compared for its performance with the classification and regression tree (CART) model and Tilt model. Database of simple declarative Kannada sentences created by Carnegie Mellon University have been deployed in this work. From the study it is very clear that F0 contours can be accurately predicted using CART and neural network models, whereas naturalness and intelligibility is high in CART model rather than neural network model.</span>