Speech Recognition Techniques Research Articles

Automatic Detection Technique for Speech Recognition based on Neural Networks Inter-Disciplinary

Automatic speech recognition allows the machine to understand and process information provided orally by a human user. It consists of using matching techniques to compare a sound wave to a set of samples, usually composed of words but also of phonemes. This field uses the knowledge of several sciences: anatomy, phonetics, signal processing, linguistics, computer science, artificial intelligence and statistics. The latest acoustic modeling methods provide deep neural networks for speech recognition. In particular, recurrent neural networks (RNNs) have several characteristics that make them a model of choice for automatic speech processing. They can keep and take into account in their decisions past and future contextual information. This paper specifically studies the behavior of Long Short-Term Memory (LSTM)-based neural networks on a specific task of automatic speech processing: speech detection. LSTM model were compared to two neural models: Multi-Layer Perceptron (MLP) and Elman’s Recurrent Neural Network (RNN). Tests on five speech detection tasks show the efficiency of the Long Short-Term Memory (LSTM) model.

Measuring and Improving Children’s Reading Aloud Attributes by Computers

Abstract In this paper, method of an automated measuring reading aloud attributes is presented. The forced alignment as a part of speech recognition technique is used. The recorded reading aloud is forced aligned to the known text and the attributes are computed from it. The tempo and fluency of children are monitored and used for an individual motivation. The length of the read text is chosen according to readers’ skills so that children end up reading at about the same time and poor readers are not frustrated. This approach has been tested and improved at the elementary school for five years and brought positive results.

Building a Smart Interactive Kiosk for Tourist Assistance

The tourism sector in morocco is increasing rapidly, is well developed, with a strong tourist industry focused on the country's coast, culture, and history. Morocco is the most politically stable countrie in North Africa, which has allowed tourism to develop. There are some tourist attractions spread over a wide geographical area, which are only visited by a few people at specific times of the year. Additionally, having human tourist guides everywhere and speaking different languages is unfeasible. This work deals with the build of a smart interactive kiosk for tourist assistance. In order to do this, it is necessary to start with an Automatic Language Recognition to recognize the language used by the tourist, then extracting keywords from what the tourist said by using Automatic Speech Recognition techniques among others, and finally the system analyses and ansers to the tourist query by suggesting a set of related informations.

Deep Elman recurrent neural networks for statistical parametric speech synthesis

Owing to the success of deep learning techniques in automatic speech recognition, deep neural networks (DNNs) have been used as acoustic models for statistical parametric speech synthesis (SPSS). DNNs do not inherently model the temporal structure in speech and text, and hence are not well suited to be directly applied to the problem of SPSS. Recurrent neural networks (RNN) on the other hand have the capability to model time-series. RNNs with long short-term memory (LSTM) cells have been shown to outperform DNN based SPSS. However, LSTM cells and its variants like gated recurrent units (GRU), simplified LSTMs (SLSTM) have complicated structure and are computationally expensive compared to the simple recurrent architecture like Elman RNN. In this paper, we explore deep Elman RNNs for SPSS and compare their effectiveness against deep gated RNNs. Specifically, we perform experiments to show that (1) Deep Elman RNNs are better suited for acoustic modeling in SPSS when compared to DNNs and perform competitively to deep SLSTMs, GRUs and LSTMs, (2) Context representation learning using Elman RNNs improves neural network acoustic models for SPSS, and (3) Elman RNN based duration model is better than the DNN based counterpart. Experiments were performed on Blizzard Challenge 2015 dataset consisting of 3 Indian languages (Telugu, Hindi and Tamil). Through subjective and objective evaluations, we show that our proposed systems outperform the baseline systems across different speakers and languages.

The Design of Secure File Transfer with Speech Recognition

The design of a secure file transfer system using Speech Recognition technique is investigated. Though cryptography enforces protection specifications, complements various other protection mechanisms. Different file types are enciphered by different cryptographic keys, supplied at access time by the users, who are being authorized using speech recognition technique. Several protection implementation schemes are analyzed according to the criteria such as: security, efficiency, and user convenience.

A comparative review of dynamic neural networks and hidden Markov model methods for mobile on-device speech recognition

The adoption of high-accuracy speech recognition algorithms without an effective evaluation of their impact on the target computational resource is impractical for mobile and embedded systems. In this paper, techniques are adopted to minimise the required computational resource for an effective mobile-based speech recognition system. A Dynamic Multi-Layer Perceptron speech recognition technique, capable of running in real time on a state-of-the-art mobile device, has been introduced. Even though a conventional hidden Markov model when applied to the same dataset slightly outperformed our approach, its processing time is much higher. The Dynamic Multi-layer Perceptron presented here has an accuracy level of 96.94% and runs significantly faster than similar techniques.

Towards automated dementia diagnosis based upon speech

The clinical diagnosis of Alzheimer’s disease and other dementias is very challenging, especially in the early stages. Our hypothesis is that any disease that effects particular brain regions involved in speech production and processing will also leave detectable finger prints in the speech. The goal of this work is an easy-to-use, non-invasive, inexpensive diagnostic test for dementia that can easily be applied in a clinician’s office or even at home. Experimental evidence suggests that strong discrimination between subjects with a diagnosis of probable Alzheimer’s versus matched normal controls can be achieved with a combination of acoustic features from speech, linguistic features extracted from a transcription of the speech, and results of a mini mental state exam. Progress is reported toward a fully automatic speech recognition system tuned for the speech-to-text aspect of this application. In addition to using state-of-the- art automatic speech recognition techniques such as Deep Learning, recurrent neural networks are used to predict the punctuation in transcribed speech, which is later used for extracting linguistic features. This fully automated system for 73 speakers is combined with 140 manually transcribed speech samples and used for experimental testing of a system for automated detection of Alzheimer’s from speech.

Analysis of Speech Disabled People with Dysarthria using TORGO Database

This paper represents a method for assessment of speech skill of dysarthria disordered patients repeatedly, which is a motor speech disorder comes from neurological injury of the motor element of the motor-speech system. This structure comprises two major features: feature representation and prediction. In the feature representation, the speech sound is changed into a phone series with an regular speech recognition technique and joint with a canonical phone sequence from an intonation dictionary using a biased finite state transducer to obtain the pronunciation mappings such as match, substitution, and deletion. Next, in the prediction process, a structured sparse linear model exists with phonological information that concurrently addresses phonologically structured sparse feature selection and transparency prediction is identified. The TORGO database of dysarthric pronunciation consists of aligned acoustical and measured 3D articulatory features from speakers with either cerebral palsy (CP) or amyotrophic lateral sclerosis (ALS), which are two of the most common causes of speech disability. The major aim of this paper is to supply a flexible and suitable environment for the dysathria speech disordered peoples by recognizing their speech to make others understand them

AN ARABIC FIGURES RECOGNITION MODEL BASED ON AUTOMATIC LEARNING OF LIP MOVEMENT

The need for an automatic speech to text conversion is continuously increasing, especially for people with special needs. Thus, automatic speech recognition techniques have been proposed to tackle such needs. The automatic recognition allows computers to identify words from lip movement, regardless of the visual source. It is known that visual speech recognition techniques improve the accuracy of word identification and shield the recognition system against acoustic noise. In this paper, we propose a hybrid voting model for automatic Arabic figures recognition based on visual perception of mouth-lip movement. The proposed model has been built for Arabic language, such that, it is able to extract Arabic figures from predefined Arabic lexicon. The predefined lexicon mainly contains the Arabic figures from zero to nine with different shapes. The model takes a video (or sequence of images) as an input, and outputs the corresponding Arabic figure for the frames extracted from the input video. Here, three techniques have been employed to extract effective visual features from mouth-lip movement. Such techniques are SURF (speeded up robust features), HoG (histogram of oriented gradient) and Haar feature extractor. The resultant features in each technique are fed separately into a classification model, namely, the hidden Markov model (HMM). The HMM identifies corresponding Arabic figure from a predefined lexicon based on input features. The final classification models that are produced from the three techniques have been grouped in a voting scheme to produce the final classification result (i.e. classification by voting). The proposed model in this paper has been tested on handcrafted data set of lip movement, and it has shown a promising result with improved accuracy of Arabic figures recognition.

A study and implementation of mapping and speech recognition techniques for an autonomous mobile robot based on ROS

Autonomous mobile robots work in close interaction with humans in environments such as homes, hospitals, public places and disaster areas. In autonomous mobile robots, the main constraints are safety, autonomy and efficiency in helping the humans. Given these constraints, developing the autonomous mobile robot technologies is a major challenge for both the industry and the research society. This paper work is about how an indoor autonomous mobile robot can work based on robot operating system and using Lidar and other sensors to create a map of an environment, and perform autonomous navigation with using capabilities like dynamic obstacle avoidance, speech recognition and video streaming. To achieve the above features, different algorithms like SLAM, AMCL, dynamic window approach algorithms, and CMU PocketSphinx speech recogniser are used. For video steaming, ROS web video server is used and the recorded video can be sent to a remote desktop system using ROS network.

A study and implementation of mapping and speech recognition techniques for an autonomous mobile robot based on ROS

Autonomous mobile robots work in close interaction with humans in environments such as homes, hospitals, public places and disaster areas. In autonomous mobile robots, the main constraints are safety, autonomy and efficiency in helping the humans. Given these constraints, developing the autonomous mobile robot technologies is a major challenge for both the industry and the research society. This paper work is about how an indoor autonomous mobile robot can work based on robot operating system and using Lidar and other sensors to create a map of an environment, and perform autonomous navigation with using capabilities like dynamic obstacle avoidance, speech recognition and video streaming. To achieve the above features, different algorithms like SLAM, AMCL, dynamic window approach algorithms, and CMU PocketSphinx speech recogniser are used. For video steaming, ROS web video server is used and the recorded video can be sent to a remote desktop system using ROS network.

A New Hybrid Intelligent Framework for Predicting Parkinson’s Disease

Parkinson’s disease (PD) is a progressive neurodegenerative motor system disorder. Early diagnosis of PD is important to control the symptoms appropriately. Recent voice and speech recognition techniques provide alternative solutions for PD screening. In this paper, an optimal support vector machine (SVM) based on bacterial foraging optimization (BFO) was established to predict PD effectively. The effectiveness of the proposed method, BFO-SVM, was validated on a PD data set based on vocal measurements. The proposed method was compared with two of the most frequently used parameter optimization methods, including an SVM based on the grid search method and an SVM based on particle swarm optimization. Additionally, to further boost the prediction accuracy, the relief feature selection was employed prior to the BFO-SVM method, consequently the RF-BFO-SVM was proposed. The experimental results have demonstrated that the proposed framework exhibited excellent classification performance with a superior classification accuracy of 97.42%.

VOICE RECOGNITION SECURITY SYSTEM USING MEL-FREQUENCY CEPSTRUM COEFFICIENTS

ABSTRACTObjective: Voice Recognition is a fascinating field spanning several areas of computer science and mathematics. Reliable speaker recognition is a hardproblem, requiring a combination of many techniques; however modern methods have been able to achieve an impressive degree of accuracy. Theobjective of this work is to examine various speech and speaker recognition techniques and to apply them to build a simple voice recognition system.Method: The project is implemented on software which uses different techniques such as Mel frequency Cepstrum Coefficient (MFCC), VectorQuantization (VQ) which are implemented using MATLAB.Results: MFCC is used to extract the characteristics from the input speech signal with respect to a particular word uttered by a particular speaker. VQcodebook is generated by clustering the training feature vectors of each speaker and then stored in the speaker database.Conclusion: Verification of the speaker is carried out using Euclidian Distance. For voice recognition we implement the MFCC approach using softwareplatform MatlabR2013b.Keywords: Mel-frequency cepstrum coefficient, Vector quantization, Voice recognition, Hidden Markov model, Euclidean distance.

SVitchboard-II and FiSVer-I: Crafting high quality and low complexity conversational english speech corpora using submodular function optimization

We introduce a set of benchmark corpora of conversational English speech derived from the Switchboard-I and Fisher datasets. Traditional automatic speech recognition (ASR) research requires considerable computational resources and has slow experimental turnaround times. Our goal is to introduce these new datasets to researchers in the ASR and machine learning communities in order to facilitate the development of novel speech recognition techniques on smaller but still acoustically rich, diverse, and hence interesting corpora. We select these corpora to maximize an acoustic quality criterion while limiting the vocabulary size (from 10 words up to 10,000 words), where both “acoustic quality” and vocabulary size are adeptly measured via various submodular functions. We also survey numerous submodular functions that could be useful to measure both “acoustic quality” and “corpus complexity” and offer guidelines on when and why a scientist may wish use to one vs. another. The corpora selection process itself is naturally performed using various state-of-the-art submodular function optimization procedures, including submodular level-set constrained submodular optimization (SCSC/SCSK), difference-of-submodular (DS) optimization, and unconstrained submodular minimization (SFM), all of which are fully defined herein. While the focus of this paper is on the resultant speech corpora, and the survey of possible objectives, a consequence of the paper is a thorough empirical comparison of the relative merits of these modern submodular optimization procedures. We provide baseline word recognition results on all of the resultant speech corpora for both Gaussian mixture model (GMM) and deep neural network (DNN)-based systems, and we have released all of the corpora definitions and Kaldi training recipes for free in the public domain.

Tamil Speech Word Recognition System with Aid of ANFIS and Dynamic Time Warping (DTW)

It is unfortunate that though the extant Tamil speech word recognition techniques have come out successful in detecting speech words from the speech word database by means of MFCC (Mel Frequency Cepstral Coefficients) features and FFBNN (Feed Forward Back Propagation Neural Network), they seem to have failed miserably to come up to expectations by generating less than desired outcomes of recital in terms of precision. Hence we have proudly launched, through this document, an innovative Tamil speech recognition technique to address the challenge by making use of novel features with ANFIS (Adaptive Neuro Fuzzy Inference System) based recognition method. Thus, at the outset, preprocessing is performed to cut down the noise in the input speech signals. Thereafter, feature vectors are mined from the preprocessed speech signals and furnished to the ANFIS. The epoch making technique is home to three novel features such as Energy Entropy, Short Time Energy and Zero Crossing Rate which are mined from the Tamil speech word signals and subjected to the word recognition procedure, in which, the ANFIS system is well guided by the features from feature mining task and the recognition efficiency is authenticated by using a set of test speech words. In the course of the testing stage, with a view to achieving exact outcomes, the dynamic time warping is estimated by means of the guidance and test word feature values. The performance outcomes illustrate the fact that the innovative Tamil speech word recognition technique has been able to achieve amazing efficiency in recognizing the input Tamil speech words, in addition to yielding higher levels of achievement in terms of precision. Moreover, the accomplishment of the well-conceived recognition technique is assessed and contrasted with the modern Tamil speech word recognition techniques.

Neural speech recognition: continuous phoneme decoding using spatiotemporal representations of human cortical activity

Objective. The superior temporal gyrus (STG) and neighboring brain regions play a key role in human language processing. Previous studies have attempted to reconstruct speech information from brain activity in the STG, but few of them incorporate the probabilistic framework and engineering methodology used in modern speech recognition systems. In this work, we describe the initial efforts toward the design of a neural speech recognition (NSR) system that performs continuous phoneme recognition on English stimuli with arbitrary vocabulary sizes using the high gamma band power of local field potentials in the STG and neighboring cortical areas obtained via electrocorticography. Approach. The system implements a Viterbi decoder that incorporates phoneme likelihood estimates from a linear discriminant analysis model and transition probabilities from an n-gram phonemic language model. Grid searches were used in an attempt to determine optimal parameterizations of the feature vectors and Viterbi decoder. Main results. The performance of the system was significantly improved by using spatiotemporal representations of the neural activity (as opposed to purely spatial representations) and by including language modeling and Viterbi decoding in the NSR system. Significance. These results emphasize the importance of modeling the temporal dynamics of neural responses when analyzing their variations with respect to varying stimuli and demonstrate that speech recognition techniques can be successfully leveraged when decoding speech from neural signals. Guided by the results detailed in this work, further development of the NSR system could have applications in the fields of automatic speech recognition and neural prosthetics.

Automatic Wheezing Detection Using Speech Recognition Technique

This study developed a speech recognition technique to detect wheezing. Wheezes are important in the diagnosis of pulmonary pathologies such as asthma. The acoustic features of wheezes are distinct in the frequency domain. Therefore, many studies have focused on detecting wheezing peaks in spectrograms through image processing. However, automated detection of wheezing peaks is difficult because of blurred edges and noise. This paper proposes an alternative approach for wheezing detection in which the mel frequency cepstral coefficients (MFCCs) are integrated into the Gaussian mixture model (GMM). The MFCCs reduce the short-term spectral information to a few coefficients, and the GMM recognizes the respiratory sounds. The respiratory sounds of 18 volunteers (9 asthmatic and 9 normal adults) were recorded for training and testing. The results of a qualitative analysis of wheeze recognition showed a good sensitivity of 0.881 and a high specificity of 0.995.

Analysis of multimodal fusion techniques for audio-visual speech recognition

Analysis of multimodal fusion techniques for audio-visual speech recognition

Establishing Efficient Security Scheme in Home IOT Devices through Biometric Finger Print Technique

Objective: To enhance the automation of home security through biometric recognition, Finger Print in home IOT. Analysis: Most of the present technologies using voice recognition technologies and the pattern of security in the home environment in which it is worked by an approved client's voice key through the confirmation procedure of Speaker Recognition and voice modulation. But this procedure may not yield optimum results when the client's speech/voice is deteriorated due to an illness or undergoing through an emotional rift. Methodology: Instead of using voice recognition technology in IOT of home security system, we opt for implement finger print recognition technique, which overcomes the problems of existing speech recognition techniques and also greatly reduces the cost of the hardware equipment. Implementation is to be done in Raspberry Pi along with the hardware modules viz., Wi-Fi router, Gas Sensor, Fire Sensor, Door Fringe motor sensor and evaluated our proposed methodology. Finding/Improvements: We implement the proposed system as a stand-alone application in Raspberry Pi2 and successful in evaluating the results which support our proposed scheme, justifying the usage of Finger Print Automation is quite efficient and economical over the usage of existing voice recognition techniques in home security automation.

Exploring the use of unsupervised query modeling techniques for speech recognition and summarization

Statistical language modeling (LM) that intends to quantify the acceptability of a given piece of text has long been an interesting yet challenging research area. In particular, language modeling for information retrieval (IR) has enjoyed remarkable empirical success; one emerging stream of the LM approach for IR is to employ the pseudo-relevance feedback process to enhance the representation of an input query so as to improve retrieval effectiveness. This paper presents a continuation of such a general line of research and the major contributions are three-fold. First, we propose a principled framework which can unify the relationships among several widely-cited query modeling formulations. Second, on top of this successfully developed framework, two extensions have been proposed. On one hand, we present an extended query modeling formulation by incorporating critical query-specific information cues to guide the model estimation. On the other hand, a word-based relevance modeling has also been leveraged to overcome the obstacle of time-consuming model estimation when the framework is being utilized for practical applications. In addition, we further adopt and formalize such a framework to the speech recognition and summarization tasks. A series of experiments reveal the empirical potential of such an LM framework and the performance merits of the deduced models on these two tasks.