Search results for: speech recognition performance

Authors: XuKe Cao

Abstract:

This paper reviews the current state of research on low-resource languages in the field of speech recognition, focusing on the challenges faced by low-resource language speech recognition, including the scarcity of data resources, the lack of linguistic resources, and the diversity of dialects and accents. The article reviews recent progress in low-resource language speech recognition, including techniques such as data augmentation, end to-end models, transfer learning, and multi-task learning. Based on the challenges currently faced, the paper also provides an outlook on future research directions. Through these studies, it is expected that the performance of speech recognition for low resource languages can be improved, promoting the widespread application and adoption of related technologies.

Keywords: low-resource languages, speech recognition, data augmentation techniques, NLP

Procedia PDF Downloads 13

Authors: João Rato, Nuno Costa

Abstract:

The human verbal communication is a two-way process which requires a mutual understanding that will result in some considerations. This kind of communication, also called dialogue, besides the supposed human agents it can also be performed between human agents and machines. The interaction between Men and Machines, by means of a natural language, has an important role concerning the improvement of the communication between each other. Aiming at knowing the performance of some speech recognition systems, this document shows the results of the accomplished tests according to the Word Error Rate evaluation method. Besides that, it is also given a set of information linked to the systems of Man-Machine communication. After this work has been made, conclusions were drawn regarding the Speech Recognition Systems, among which it can be mentioned their poor performance concerning the voice interpretation in noisy environments.

Keywords: automatic speech recognition, man-machine conversation, speech recognition, spoken dialogue systems, word error rate

Procedia PDF Downloads 322

Authors: Khaled M. Alhawiti

Abstract:

This research study aims to present a retrospective study about speech recognition systems and artificial intelligence. Speech recognition has become one of the widely used technologies, as it offers great opportunity to interact and communicate with automated machines. Precisely, it can be affirmed that speech recognition facilitates its users and helps them to perform their daily routine tasks, in a more convenient and effective manner. This research intends to present the illustration of recent technological advancements, which are associated with artificial intelligence. Recent researches have revealed the fact that speech recognition is found to be the utmost issue, which affects the decoding of speech. In order to overcome these issues, different statistical models were developed by the researchers. Some of the most prominent statistical models include acoustic model (AM), language model (LM), lexicon model, and hidden Markov models (HMM). The research will help in understanding all of these statistical models of speech recognition. Researchers have also formulated different decoding methods, which are being utilized for realistic decoding tasks and constrained artificial languages. These decoding methods include pattern recognition, acoustic phonetic, and artificial intelligence. It has been recognized that artificial intelligence is the most efficient and reliable methods, which are being used in speech recognition.

Keywords: speech recognition, acoustic phonetic, artificial intelligence, hidden markov models (HMM), statistical models of speech recognition, human machine performance

Procedia PDF Downloads 478

Authors: Van Nhan Nguyen, Harald Holone

Abstract:

Over the past few years, a lot of research has been conducted to bring Automatic Speech Recognition (ASR) into various areas of Air Traffic Control (ATC), such as air traffic control simulation and training, monitoring live operators for with the aim of safety improvements, air traffic controller workload measurement and conducting analysis on large quantities controller-pilot speech. Due to the high accuracy requirements of the ATC context and its unique challenges, automatic speech recognition has not been widely adopted in this field. With the aim of providing a good starting point for researchers who are interested bringing automatic speech recognition into ATC, this paper gives an overview of possibilities and challenges of applying automatic speech recognition in air traffic control. To provide this overview, we present an updated literature review of speech recognition technologies in general, as well as specific approaches relevant to the ATC context. Based on this literature review, criteria for selecting speech recognition approaches for the ATC domain are presented, and remaining challenges and possible solutions are discussed.

Keywords: automatic speech recognition, asr, air traffic control, atc

Procedia PDF Downloads 399

Authors: Fawaz S. Al-Anzi, Dia AbuZeina

Abstract:

Speech recognition is of an important contribution in promoting new technologies in human computer interaction. Today, there is a growing need to employ speech technology in daily life and business activities. However, speech recognition is a challenging task that requires different stages before obtaining the desired output. Among automatic speech recognition (ASR) components is the feature extraction process, which parameterizes the speech signal to produce the corresponding feature vectors. Feature extraction process aims at approximating the linguistic content that is conveyed by the input speech signal. In speech processing field, there are several methods to extract speech features, however, Mel Frequency Cepstral Coefficients (MFCC) is the popular technique. It has been long observed that the MFCC is dominantly used in the well-known recognizers such as the Carnegie Mellon University (CMU) Sphinx and the Markov Model Toolkit (HTK). Hence, this paper focuses on the MFCC method as the standard choice to identify the different speech segments in order to obtain the language phonemes for further training and decoding steps. Due to MFCC good performance, the previous studies show that the MFCC dominates the Arabic ASR research. In this paper, we demonstrate MFCC as well as the intermediate steps that are performed to get these coefficients using the HTK toolkit.

Keywords: speech recognition, acoustic features, mel frequency, cepstral coefficients

Procedia PDF Downloads 259

Authors: S. Jagadeesh Kumar

Abstract:

This expose presents a luculent of recent machine learning practices as employed in the modern and as pertinent to prospective automatic speech recognition schemes. The aspiration is to promote additional traverse ablution among the machine learning and automatic speech recognition factions that have transpired in the precedent. The manuscript is structured according to the chief machine learning archetypes that are furthermore trendy by now or have latency for building momentous hand-outs to automatic speech recognition expertise. The standards offered and convoluted in this article embraces adaptive and multi-task learning, active learning, Bayesian learning, discriminative learning, generative learning, supervised and unsupervised learning. These learning archetypes are aggravated and conferred in the perspective of automatic speech recognition tools and functions. This manuscript bequeaths and surveys topical advances of deep learning and learning with sparse depictions; further limelight is on their incessant significance in the evolution of automatic speech recognition.

Keywords: automatic speech recognition, deep learning methods, machine learning archetypes, Bayesian learning, supervised and unsupervised learning

Procedia PDF Downloads 447

Authors: S. B. Rathna Kumar, Pranjali A Ujwane, Panchanan Mohanty

Abstract:

The present study aimed to develop a battery for assessing speech recognition performance by adults in Marathi. A total of four word lists were developed by considering word frequency, word familiarity, words in common use, and phonemic balance. Each word list consists of 25 words (15 monosyllabic words in CVC structure and 10 monosyllabic words in CVCV structure). Equivalence analysis and performance-intensity function testing was carried using the four word lists on a total of 150 native speakers of Marathi belonging to different regions of Maharashtra (Vidarbha, Marathwada, Khandesh and Northern Maharashtra, Pune, and Konkan). The subjects were further equally divided into five groups based on above mentioned regions. It was found that there was no significant difference (p > 0.05) in the speech recognition performance between groups for each word list and between word lists for each group. Hence, the four word lists developed were equally difficult for all the groups and can be used interchangeably. The performance-intensity (PI) function curve showed semi-linear function, and the groups’ mean slope of the linear portions of the curve indicated an average linear slope of 4.64%, 4.73%, 4.68%, and 4.85% increase in word recognition score per dB for list 1, list 2, list 3 and list 4 respectively. Although, there is no data available on speech recognition tests for adults in Marathi, most of the findings of the study are in line with the findings of research reports on other languages. The four word lists, thus developed, were found to have sufficient reliability and validity in assessing speech recognition performance by adults in Marathi.

Keywords: speech recognition performance, phonemic balance, equivalence analysis, performance-intensity function testing, reliability, validity

Procedia PDF Downloads 356

Authors: R. Ajgou, S. Sbaa, S. Ghendir, A. Chemsa, A. Taleb-Ahmed

Abstract:

The speech enhancement algorithm is to improve speech quality. In this paper, we review some speech enhancement methods and we evaluated their performance based on Perceptual Evaluation of Speech Quality scores (PESQ, ITU-T P.862). All method was evaluated in presence of different kind of noise using TIMIT database and NOIZEUS noisy speech corpus.. The noise was taken from the AURORA database and includes suburban train noise, babble, car, exhibition hall, restaurant, street, airport and train station noise. Simulation results showed improved performance of speech enhancement for Tracking of non-stationary noise approach in comparison with various methods in terms of PESQ measure. Moreover, we have evaluated the effects of the speech enhancement technique on Speaker Identification system based on autoregressive (AR) model and Mel-frequency Cepstral coefficients (MFCC).

Keywords: speech enhancement, pesq, speaker recognition, MFCC

Procedia PDF Downloads 424

Authors: Mumtaz Begum Mustafa, Siti Salwah Salim, Feizal Dani Rahman

Abstract:

Recently, Automatic Speech Recognition (ASR) systems were used to assist children in language acquisition as it has the ability to detect human speech signal. Despite the benefits offered by the ASR system, there is a lack of ASR systems for Malay-speaking children. One of the contributing factors for this is the lack of continuous speech database for the target users. Though cross-lingual adaptation is a common solution for developing ASR systems for under-resourced language, it is not viable for children as there are very limited speech databases as a source model. In this research, we propose a two-stage adaptation for the development of ASR system for Malay-speaking children using a very limited database. The two stage adaptation comprises the cross-lingual adaptation (first stage) and cross-age adaptation. For the first stage, a well-known speech database that is phonetically rich and balanced, is adapted to the medium-sized Malay adults using supervised MLLR. The second stage adaptation uses the speech acoustic model generated from the first adaptation, and the target database is a small-sized database of the target users. We have measured the performance of the proposed technique using word error rate, and then compare them with the conventional benchmark adaptation. The two stage adaptation proposed in this research has better recognition accuracy as compared to the benchmark adaptation in recognizing children’s speech.

Keywords: Automatic Speech Recognition System, children speech, adaptation, Malay

Procedia PDF Downloads 397

Authors: Bubai Maji, Monorama Swain

Abstract:

Speech is considered an essential and most natural medium for the interaction between machines and humans. However, extracting effective features for speech emotion recognition (SER) is remains challenging. The present studies show that the temporal information captured but high-level temporal-feature learning is yet to be investigated. In this paper, we present an efficient novel method using the Self-attention (SA) mechanism in a combination of Convolutional Neural Network (CNN) and Bi-directional Gated Recurrent Unit (Bi-GRU) network to learn high-level temporal-feature. In order to further enhance the representation of the high-level temporal-feature, we integrate a Bi-GRU output with learnable weights features by SA, and improve the performance. We evaluate our proposed method on our created SITB-OSED and IEMOCAP databases. We report that the experimental results of our proposed method achieve state-of-the-art performance on both databases.

Keywords: Bi-GRU, 1D-CNNs, self-attention, speech emotion recognition

Procedia PDF Downloads 113

Authors: Khenfer-Koummich Fatma, Hendel Fatiha, Mesbahi Larbi

Abstract:

this paper presents an approach based on Hidden Markov Models (HMM: Hidden Markov Model) using HTK tools. The goal is to create a man-machine interface with a voice recognition system that allows the operator to tele-operate a mentor robot to execute specific tasks as rotate, raise, close, etc. This system should take into account different levels of environmental noise. This approach has been applied to isolated words representing the robot commands spoken in two languages: French and Arabic. The recognition rate obtained is the same in both speeches, Arabic and French in the neutral words. However, there is a slight difference in favor of the Arabic speech when Gaussian white noise is added with a Signal to Noise Ratio (SNR) equal to 30 db, the Arabic speech recognition rate is 69% and 80% for French speech recognition rate. This can be explained by the ability of phonetic context of each speech when the noise is added.

Keywords: voice command, HMM, TIMIT, noise, HTK, Arabic, speech recognition

Procedia PDF Downloads 382

Authors: Khenfer Koummich Fatma, Hendel Fatiha, Mesbahi Larbi

Abstract:

This paper presents an approach based on Hidden Markov Models (HMM: Hidden Markov Model) using HTK tools. The goal is to create a human-machine interface with a voice recognition system that allows the operator to teleoperate a mentor robot to execute specific tasks as rotate, raise, close, etc. This system should take into account different levels of environmental noise. This approach has been applied to isolated words representing the robot commands pronounced in two languages: French and Arabic. The obtained recognition rate is the same in both speeches, Arabic and French in the neutral words. However, there is a slight difference in favor of the Arabic speech when Gaussian white noise is added with a Signal to Noise Ratio (SNR) equals 30 dB, in this case; the Arabic speech recognition rate is 69%, and the French speech recognition rate is 80%. This can be explained by the ability of phonetic context of each speech when the noise is added.

Keywords: Arabic speech recognition, Hidden Markov Model (HMM), HTK, noise, TIMIT, voice command

Procedia PDF Downloads 385

Authors: John Lorenzo Bautista, Yoon-Joong Kim

Abstract:

This paper presents the development of a Filipino speech recognition tool using the HTK System. The system was trained from a subset of the Filipino Speech Corpus developed by the DSP Laboratory of the University of the Philippines-Diliman. The speech corpus was both used in training and testing the system by estimating the parameters for phonetic HMM-based (Hidden-Markov Model) acoustic models. Experiments on different mixture-weights were incorporated in the study. The phoneme-level word-based recognition of a 5-state HMM resulted in an average accuracy rate of 80.13 for a single-Gaussian mixture model, 81.13 after implementing a phoneme-alignment, and 87.19 for the increased Gaussian-mixture weight model. The highest accuracy rate of 88.70% was obtained from a 5-state model with 6 Gaussian mixtures.

Keywords: Filipino language, Hidden Markov Model, HTK system, speech recognition

Procedia PDF Downloads 480

Authors: G. Tamulevičius, A. Serackis, T. Sledevič, D. Navakauskas

Abstract:

We consider the biggest challenge in speech recognition – noise reduction. Traditionally detected transient noise pulses are removed with the corrupted speech using pulse models. In this paper we propose to cope with the problem directly in Dynamic Time Warping domain. Bidirectional Dynamic Time Warping algorithm for the recognition of isolated words impacted by transient noise pulses is proposed. It uses simple transient noise pulse detector, employs bidirectional computation of dynamic time warping and directly manipulates with warping results. Experimental investigation with several alternative solutions confirms effectiveness of the proposed algorithm in the reduction of impact of noise on recognition process – 3.9% increase of the noisy speech recognition is achieved.

Keywords: transient noise pulses, noise reduction, dynamic time warping, speech recognition

Procedia PDF Downloads 558

Authors: A. Shoiynbek, K. Kozhakhmet, P. Menezes, D. Kuanyshbay, D. Bayazitov

Abstract:

Speech emotion recognition has received increasing research interest all through current years. There was used emotional speech that was collected under controlled conditions in most research work. Actors imitating and artificially producing emotions in front of a microphone noted those records. There are four issues related to that approach, namely, (1) emotions are not natural, and it means that machines are learning to recognize fake emotions. (2) Emotions are very limited by quantity and poor in their variety of speaking. (3) There is language dependency on SER. (4) Consequently, each time when researchers want to start work with SER, they need to find a good emotional database on their language. In this paper, we propose the approach to create an automatic tool for speech emotion extraction based on facial emotion recognition and describe the sequence of actions of the proposed approach. One of the first objectives of the sequence of actions is a speech detection issue. The paper gives a detailed description of the speech detection model based on a fully connected deep neural network for Kazakh and Russian languages. Despite the high results in speech detection for Kazakh and Russian, the described process is suitable for any language. To illustrate the working capacity of the developed model, we have performed an analysis of speech detection and extraction from real tasks.

Keywords: deep neural networks, speech detection, speech emotion recognition, Mel-frequency cepstrum coefficients, collecting speech emotion corpus, collecting speech emotion dataset, Kazakh speech dataset

Procedia PDF Downloads 101

Authors: Yunbin Deng

Abstract:

Most existing applications of automatic speech recognition relies on cooperative subjects at a short distance to a microphone. Standoff speech recognition using microphone arrays can extend the subject to sensor distance somewhat, but it is still limited to only a few feet. As such, most deployed applications of standoff speech recognitions are limited to indoor use at short range. Moreover, these applications require air passway between the subject and the sensor to achieve reasonable signal to noise ratio. This study reports long range (50 feet) automatic speech recognition experiments using a Laser Doppler Vibrometer (LDV) sensor. This study shows that the LDV sensor modality can extend the speech acquisition standoff distance far beyond microphone arrays to hundreds of feet. In addition, LDV enables 'listening' through the windows for uncooperative subjects. This enables new capabilities in automatic audio and speech intelligence, surveillance, and reconnaissance (ISR) for law enforcement, homeland security and counter terrorism applications. The Polytec LDV model OFV-505 is used in this study. To investigate the impact of different vibrating materials, five parallel LDV speech corpora, each consisting of 630 speakers, are collected from the vibrations of a glass window, a metal plate, a plastic box, a wood slate, and a concrete wall. These are the common materials the application could encounter in a daily life. These data were compared with the microphone counterpart to manifest the impact of various materials on the spectrum of the LDV speech signal. State of the art deep neural network modeling approaches is used to conduct continuous speaker independent speech recognition on these LDV speech datasets. Preliminary phoneme recognition results using time-delay neural network, bi-directional long short term memory, and model fusion shows great promise of using LDV for long range speech recognition. To author’s best knowledge, this is the first time an LDV is reported for long distance speech recognition application.

Keywords: covert speech acquisition, distant speech recognition, DSR, laser Doppler vibrometer, LDV, speech intelligence surveillance and reconnaissance, ISR

Procedia PDF Downloads 179

Authors: Aisultan Shoiynbek, Darkhan Kuanyshbay, Paulo Menezes, Akbayan Bekarystankyzy, Assylbek Mukhametzhanov, Temirlan Shoiynbek

Abstract:

Speech emotion recognition (SER) has received increasing research interest in recent years. It is a common practice to utilize emotional speech collected under controlled conditions recorded by actors imitating and artificially producing emotions in front of a microphone. There are four issues related to that approach: emotions are not natural, meaning that machines are learning to recognize fake emotions; emotions are very limited in quantity and poor in variety of speaking; there is some language dependency in SER; consequently, each time researchers want to start work with SER, they need to find a good emotional database in their language. This paper proposes an approach to create an automatic tool for speech emotion extraction based on facial emotion recognition and describes the sequence of actions involved in the proposed approach. One of the first objectives in the sequence of actions is the speech detection issue. The paper provides a detailed description of the speech detection model based on a fully connected deep neural network for Kazakh and Russian. Despite the high results in speech detection for Kazakh and Russian, the described process is suitable for any language. To investigate the working capacity of the developed model, an analysis of speech detection and extraction from real tasks has been performed.

Keywords: deep neural networks, speech detection, speech emotion recognition, Mel-frequency cepstrum coefficients, collecting speech emotion corpus, collecting speech emotion dataset, Kazakh speech dataset

Procedia PDF Downloads 26

Authors: Brahim Fares Zaidi

Abstract:

Our work aims to improve our Automatic Recognition System for Dysarthria Speech based on the Hidden Models of Markov and the Hidden Markov Model Toolkit to help people who are sick. With pronunciation problems, we applied two techniques of speech parameterization based on Mel Frequency Cepstral Coefficients and Perceptual Linear Prediction and concatenated them with JITTER and SHIMMER coefficients in order to increase the recognition rate of a dysarthria speech. For our tests, we used the NEMOURS database that represents speakers with dysarthria and normal speakers.

Keywords: ARSDS, HTK, HMM, MFCC, PLP

Procedia PDF Downloads 108

Authors: Pogula Rakesh, T. Kishore Kumar

Abstract:

Speech enhancement is a long standing problem with numerous applications like teleconferencing, VoIP, hearing aids, and speech recognition. The motivation behind this research work is to obtain a clean speech signal of higher quality by applying the optimal noise cancellation technique. Real-time adaptive filtering algorithms seem to be the best candidate among all categories of the speech enhancement methods. In this paper, we propose a speech enhancement method based on Recursive Least Squares (RLS) adaptive filter of speech signals. Experiments were performed on noisy data which was prepared by adding AWGN, Babble and Pink noise to clean speech samples at -5dB, 0dB, 5dB, and 10dB SNR levels. We then compare the noise cancellation performance of proposed RLS algorithm with existing NLMS algorithm in terms of Mean Squared Error (MSE), Signal to Noise ratio (SNR), and SNR loss. Based on the performance evaluation, the proposed RLS algorithm was found to be a better optimal noise cancellation technique for speech signals.

Keywords: adaptive filter, adaptive noise canceller, mean squared error, noise reduction, NLMS, RLS, SNR, SNR loss

Procedia PDF Downloads 481

Authors: Jong Han Joo, Jung Hoon Lee, Young Sun Kim, Jae Young Kang, Seung Ho Choi

Abstract:

In this study, we propose a novel technique for acoustic echo suppression (AES) during speech recognition under barge-in conditions. Conventional AES methods based on spectral subtraction apply fixed weights to the estimated echo path transfer function (EPTF) at the current signal segment and to the EPTF estimated until the previous time interval. We propose a new approach that adaptively updates weight parameters in response to abrupt changes in the acoustic environment due to background noises or double-talk. Furthermore, we devised a voice activity detector and an initial time-delay estimator for barge-in speech recognition in communication networks. The initial time delay is estimated using log-spectral distance measure, as well as cross-correlation coefficients. The experimental results show that the developed techniques can be successfully applied in barge-in speech recognition systems.

Keywords: acoustic echo suppression, barge-in, speech recognition, echo path transfer function, initial delay estimator, voice activity detector

Procedia PDF Downloads 372

Authors: M. Draman, S. Z. Muhamad Yassin, M. S. Alias, Z. Lambak, M. I. Zulkifli, S. N. Padhi, K. N. Baharim, F. Maskuriy, A. I. A. Rahim

Abstract:

The performance of Malay automatic speech recognition (ASR) system for the call centre environment is presented. The system utilizes Kaldi toolkit as the platform to the entire library and algorithm used in performing the ASR task. The acoustic model implemented in this system uses a deep neural network (DNN) method to model the acoustic signal and the standard (n-gram) model for language modelling. With 80 hours of training data from the call centre recordings, the ASR system can achieve 72% of accuracy that corresponds to 28% of word error rate (WER). The testing was done using 20 hours of audio data. Despite the implementation of DNN, the system shows a low accuracy owing to the varieties of noises, accent and dialect that typically occurs in Malaysian call centre environment. This significant variation of speakers is reflected by the large standard deviation of the average word error rate (WERav) (i.e., ~ 10%). It is observed that the lowest WER (13.8%) was obtained from recording sample with a standard Malay dialect (central Malaysia) of native speaker as compared to 49% of the sample with the highest WER that contains conversation of the speaker that uses non-standard Malay dialect.

Keywords: conversational speech recognition, deep neural network, Malay language, speech recognition

Procedia PDF Downloads 322

Authors: Deawan Rakin Ahamed Remal, Sinthia Chowdhury, Sharun Akter Khushbu, Sheak Rashed Haider Noori

Abstract:

Native Speech to text synthesis has its own leverage for the purpose of mankind. The extensive nature of art to speaking different accents is common but the purpose of communication between two different accent types of people is quite difficult. This problem will be motivated by the extraction of the wrong perception of language meaning. Thus, many existing automatic speech recognition has been placed to detect text. Overall study of this paper mentions a review of NSTTR (Native Speech Text to Text Recognition) synthesis compared with Text to Text recognition. Review has exposed many text to text recognition systems that are at a very early stage to comply with the system by native speech recognition. Many discussions started about the progression of chatbots, linguistic theory another is rule based approach. In the Recent years Deep learning is an overwhelming chapter for text to text learning to detect language nature. To the best of our knowledge, In the sub continent a huge number of people speak in Bangla language but they have different accents in different regions therefore study has been elaborate contradictory discussion achievement of existing works and findings of future needs in Bangla language acoustic accent.

Keywords: TTR, NSTTR, text to text recognition, deep learning, natural language processing

Procedia PDF Downloads 132

Authors: E. Krasnova, E. Bulgakova, V. Shchemelinin

Abstract:

The paper deals with acoustic-spectrographic voice identification method in terms of its performance in non-native language speech. Performance evaluation is conducted by comparing the result of the analysis of recordings containing native language speech with recordings that contain foreign language speech. Our research is based on Tajik and Russian speech of Tajik native speakers due to the character of the criminal situation with drug trafficking. We propose a pilot experiment that represents a primary attempt enter the field.

Keywords: speaker identification, acoustic-spectrographic method, non-native speech, performance evaluation

Procedia PDF Downloads 446

Authors: Bipasha Sen, Aditya Agarwal

Abstract:

Multilingual automatic speech recognition (ASR) system is a single entity capable of transcribing multiple languages sharing a common phone space. Performance of such a system is highly dependent on the compatibility of the languages. State of the art speech recognition systems are built using sequential architectures based on recurrent neural networks (RNN) limiting the computational parallelization in training. This poses a significant challenge in terms of time taken to bootstrap and validate the compatibility of multiple languages for building a robust multilingual system. Complex architectural choices based on self-attention networks are made to improve the parallelization thereby reducing the training time. In this work, we propose Reed, a simple system based on 1D convolutions which uses very short context to improve the training time. To improve the performance of our system, we use raw time-domain speech signals directly as input. This enables the convolutional layers to learn feature representations rather than relying on handcrafted features such as MFCC. We report improvement on training and inference times by atleast a factor of 4x and 7.4x respectively with comparable WERs against standard RNN based baseline systems on SpeechOcean's multilingual low resource dataset.

Keywords: convolutional neural networks, language compatibility, low resource languages, multilingual automatic speech recognition

Procedia PDF Downloads 123

Authors: Jalal Karam

Abstract:

The methods of Fourier, Laplace, and Wavelet Transforms provide transfer functions and relationships between the input and the output signals in linear time invariant systems. This paper shows the equivalence among these three methods and in each case presenting an application of the appropriate (Fourier, Laplace or Wavelet) to the convolution theorem. In addition, it is shown that the same holds for a direct integration method. The Biorthogonal wavelets Bior3.5 and Bior3.9 are examined and the zeros distribution of their polynomials associated filters are located. This paper also presents the significance of utilizing wavelets as effective tools in processing speech signals for common multimedia applications in general, and for recognition and compression in particular. Theoretically and practically, wavelets have proved to be effective and competitive. The practical use of the Continuous Wavelet Transform (CWT) in processing and analysis of speech is then presented along with explanations of how the human ear can be thought of as a natural wavelet transformer of speech. This generates a variety of approaches for applying the (CWT) to many paradigms analysing speech, sound and music. For perception, the flexibility of implementation of this transform allows the construction of numerous scales and we include two of them. Results for speech recognition and speech compression are then included.

Keywords: continuous wavelet transform, biorthogonal wavelets, speech perception, recognition and compression

Procedia PDF Downloads 416

Authors: Elena E. Lyakso, Olga V. Frolova, Yuri N. Matveev, Aleksey S. Grigorev, Alexander S. Nikolaev, Viktor A. Gorodnyi

Abstract:

The study of the children’s emotional sphere depending on age and psychoneurological state is of great importance for the design of educational programs for children and their social adaptation. Atypical development may be accompanied by violations or specificities of the emotional sphere. To study characteristics of the emotional state reflection in the voice and speech features of children, the perceptual study with the participation of adults and the automatic recognition of speech were conducted. Speech of children with typical development (TD), with Down syndrome (DS), and with autism spectrum disorders (ASD) aged 6-12 years was recorded. To obtain emotional speech in children, model situations were created, including a dialogue between the child and the experimenter containing questions that can cause various emotional states in the child and playing with a standard set of toys. The questions and toys were selected, taking into account the child’s age, developmental characteristics, and speech skills. For the perceptual experiment by adults, test sequences containing speech material of 30 children: TD, DS, and ASD were created. The listeners were 100 adults (age 19.3 ± 2.3 years). The listeners were tasked with determining the children’s emotional state as “comfort – neutral – discomfort” while listening to the test material. Spectrographic analysis of speech signals was conducted. For automatic recognition of the emotional state, 6594 speech files containing speech material of children were prepared. Automatic recognition of three states, “comfort – neutral – discomfort,” was performed using automatically extracted from the set of acoustic features - the Geneva Minimalistic Acoustic Parameter Set (GeMAPS) and the extended Geneva Minimalistic Acoustic Parameter Set (eGeMAPS). The results showed that the emotional state is worse determined by the speech of TD children (comfort – 58% of correct answers, discomfort – 56%). Listeners better recognized discomfort in children with ASD and DS (78% of answers) than comfort (70% and 67%, respectively, for children with DS and ASD). The neutral state is better recognized by the speech of children with ASD (67%) than by the speech of children with DS (52%) and TD children (54%). According to the automatic recognition data using the acoustic feature set GeMAPSv01b, the accuracy of automatic recognition of emotional states for children with ASD is 0.687; children with DS – 0.725; TD children – 0.641. When using the acoustic feature set eGeMAPSv01b, the accuracy of automatic recognition of emotional states for children with ASD is 0.671; children with DS – 0.717; TD children – 0.631. The use of different models showed similar results, with better recognition of emotional states by the speech of children with DS than by the speech of children with ASD. The state of comfort is automatically determined better by the speech of TD children (precision – 0.546) and children with ASD (0.523), discomfort – children with DS (0.504). The data on the specificities of recognition by adults of the children’s emotional state by their speech may be used in recruitment for working with children with atypical development. Automatic recognition data can be used to create alternative communication systems and automatic human-computer interfaces for social-emotional learning. Acknowledgment: This work was financially supported by the Russian Science Foundation (project 18-18-00063).

Keywords: autism spectrum disorders, automatic recognition of speech, child’s emotional speech, Down syndrome, perceptual experiment

Procedia PDF Downloads 189

Authors: Hadeer M. Sayed, Hesham E. El Deeb, Shereen A. Taie

Abstract:

In the big data era, we are facing a diversity of datasets from different sources in different domains that describe a single life event. These datasets consist of multiple modalities, each of which has a different representation, distribution, scale, and density. Multimodal fusion is the concept of integrating information from multiple modalities in a joint representation with the goal of predicting an outcome through a classification task or regression task. In this paper, multimodal fusion techniques are classified into two main classes: model-agnostic techniques and model-based approaches. It provides a comprehensive study of recent research in each class and outlines the benefits and limitations of each of them. Furthermore, the audiovisual speech recognition task is expressed as a case study of multimodal data fusion approaches, and the open issues through the limitations of the current studies are presented. This paper can be considered a powerful guide for interested researchers in the field of multimodal data fusion and audiovisual speech recognition particularly.

Keywords: multimodal data, data fusion, audio-visual speech recognition, neural networks

Procedia PDF Downloads 111

Authors: Tomasz Sapiński, Dorota Kamińska, Adam Pelikant, Egils Avots, Cagri Ozcinar, Gholamreza Anbarjafari

Abstract:

People express emotions through different modalities. Integration of verbal and non-verbal communication channels creates a system in which the message is easier to understand. Expanding the focus to several expression forms can facilitate research on emotion recognition as well as human-machine interaction. In this article, the authors present a Polish emotional database composed of three modalities: facial expressions, body movement and gestures, and speech. The corpora contains recordings registered in studio conditions, acted out by 16 professional actors (8 male and 8 female). The data is labeled with six basic emotions categories, according to Ekman’s emotion categories. To check the quality of performance, all recordings are evaluated by experts and volunteers. The database is available to academic community and might be useful in the study on audio-visual emotion recognition.

Keywords: body movement, emotion recognition, emotional corpus, facial expressions, gestures, multimodal database, speech

Procedia PDF Downloads 349

Authors: Hadri Hussain, M.Nasir Ibrahim, Chee-Ming Ting, Mariani Idroas, Fuad Numan, Alias Mohd Noor

Abstract:

A biometric security system has become an important application in client identification and verification system. A conventional biometric system is normally based on unimodal biometric that depends on either behavioural or physiological information for authentication purposes. The behavioural biometric depends on human body biometric signal (such as speech) and biosignal biometric (such as electrocardiogram (ECG) and phonocardiogram or heart sound (HS)). The speech signal is commonly used in a recognition system in biometric, while the ECG and the HS have been used to identify a person’s diseases uniquely related to its cluster. However, the conventional biometric system is liable to spoof attack that will affect the performance of the system. Therefore, a multimodal biometric security system is developed, which is based on biometric signal of ECG, HS, and speech. The biosignal data involved in the biometric system is initially segmented, with each segment Mel Frequency Cepstral Coefficients (MFCC) method is exploited for extracting the feature. The Hidden Markov Model (HMM) is used to model the client and to classify the unknown input with respect to the modal. The recognition system involved training and testing session that is known as client identification (CID). In this project, twenty clients are tested with the developed system. The best overall performance at 44 kHz was 93.92% for ECG and the worst overall performance was ECG at 88.47%. The results were compared to the best overall performance at 44 kHz for (20clients) to increment of clients, which was 90.00% for HS and the worst overall performance falls at ECG at 79.91%. It can be concluded that the difference multimodal biometric has a substantial effect on performance of the biometric system and with the increment of data, even with higher frequency sampling, the performance still decreased slightly as predicted.

Keywords: electrocardiogram, phonocardiogram, hidden markov model, mel frequency cepstral coeffiecients, client identification

Procedia PDF Downloads 280

Authors: Shaode Yu, Jiajian Meng, Bing Zhu, Hang Yu, Qiurui Sun

Abstract:

Speech emotion recognition (SER) is for recognizing human subjective emotions through audio data in-depth analysis. From speech audios, how to comprehensively extract emotional information and how to effectively fuse extracted features remain challenging. This paper presents a dual-stream SER framework that embraces both full training and transfer learning of different networks for thorough feature encoding. Besides, a plug-and-play cross-attention fusion (CAF) module is implemented for the valid integration of the dual-stream encoder output. The effectiveness of the proposed CAF module is compared to the other three fusion modules (feature summation, feature concatenation, and feature-wise linear modulation) on two databases (RAVDESS and IEMO-CAP) using different dual-stream encoders (full training network, DPCNN or TextRCNN; transfer learning network, HuBERT or Wav2Vec2). Experimental results suggest that the CAF module can effectively reconcile conflicts between features from different encoders and outperform the other three feature fusion modules on the SER task. In the future, the plug-and-play CAF module can be extended for multi-branch feature fusion, and the dual-stream SER framework can be widened for multi-stream data representation to improve the recognition performance and generalization capacity.

Keywords: speech emotion recognition, cross-attention fusion, dual-stream, pre-trained

Procedia PDF Downloads 75