Speech Encoding Research Articles

A robust cryptosystem to enhance the security in speech based person authentication

The developments in technology have made us utilizing speech as a biometric to authenticate persons. In this paper, speech encryption and decryption algorithm are presented for enhancing the security in speech-based person authentication systems. The implementation of the authentication system contains the feature extraction, modeling techniques and testing procedures for authenticating the person. Firstly, the Mel frequency cepstral coefficient (MFCC) features are extracted from the training speech utterances and models are developed for each speaker. The speech encryption system encrypts the test speech utterances. Multiple chaotic mapping techniques and Deoxyribonucleic acid (DNA) addition based speech cryptosystem is developed to secure test speech against attacks. The speech encryption system deals with sampled test speech signal given as input, which is subjected to intra level and inter level bit substitution. These resultant samples are encoded into the DNA sequence denoted by P(n). The DNA sequence P(n) and DNA sequences {A(n), B(n), C(n), D(n)} obtained using different techniques based on chaos, such as tent mapping, henon mapping, sine mapping, and logistic mapping and summed up together using DNA addition operation. Finally, the encrypted test speech is obtained using DNA decoding. The speaker authentication system in the receiving side decrypts the encrypted signal and identifies the speakers from the decrypted speech. The correlation coefficient test, Signal to noise ratio test, Peak Signal to Noise Ratio test, key sensitivity test, NSCR and UACI test, key space analysis, and histogram analysis are the techniques used as metrics to prove the efficiency of the proposed cryptosystem. Overall individual accuracy is 97% for the text dependent person authentication with the original test speech set and decrypted test speech set. Overall individual accuracy is 66% for the text independent person authentication with the original test speech set and decrypted test speech set. In our work, the speech utterances are taken from AVSpoof database for authenticating 44 speakers. Our work highlights the efficiency of the encryption system, to provide security for test speech and person authentication using speech as a biometric.

Bridging the Gap between Pre-Training and Fine-Tuning for End-to-End Speech Translation

End-to-end speech translation, a hot topic in recent years, aims to translate a segment of audio into a specific language with an end-to-end model. Conventional approaches employ multi-task learning and pre-training methods for this task, but they suffer from the huge gap between pre-training and fine-tuning. To address these issues, we propose a Tandem Connectionist Encoding Network (TCEN) which bridges the gap by reusing all subnets in fine-tuning, keeping the roles of subnets consistent, and pre-training the attention module. Furthermore, we propose two simple but effective methods to guarantee the speech encoder outputs and the MT encoder inputs are consistent in terms of semantic representation and sequence length. Experimental results show that our model leads to significant improvements in En-De and En-Fr translation irrespective of the backbones.

Hardware realization of a secure and enhanced s-box based speech encryption engine

This paper presents a secure and efficient substitution box (s-box) for speech encryption applications. The proposed s-box data changes every clock cycle to swap the input signal with different data, where it generated based on a new algorithm and a memristor chaotic system. Bifurcation diagrams for all memristor chaotic system parameters are introduced to stand for the chaotic range of each parameter. Moreover, the effect of each component inside the proposed encryption system is studied, and the security of the system is validated through perceptual and statistical tests. The size of the encryption key is 175 bits to meet the global standards for the optimum encryption key width (> 128). MATLAB software is used to calculate entropy, MSE, and correlation coefficient. Both chaotic circuit and encryption/decryption schemes are designed using Verilog HDL and simulated by Xilinx ISE 14.7. Xilinx Virtex 5 FPGA kit is used to realize the proposed algorithm with a throughput 0.536 of Gbit/s. The cryptosystem is tested using two different speech files to examine its efficiency.

Direct Speech-to-Image Translation

Direct speech-to-image translation without text is an interesting and useful topic due to the potential applications in human-computer interaction, art creation, computer-aided design. etc. Not to mention that many languages have no writing form. However, as far as we know, it has not been well-studied how to translate the speech signals into images directly and how well they can be translated. In this paper, we attempt to translate the speech signals into the image signals without the transcription stage. Specifically, a speech encoder is designed to represent the input speech signals as an embedding feature, and it is trained with a pretrained image encoder using teacher-student learning to obtain better generalization ability on new classes. Subsequently, a stacked generative adversarial network is used to synthesize high-quality images conditioned on the embedding feature. Experimental results on both synthesized and real data show that our proposed method is effective to translate the raw speech signals into images without the middle text representation. Ablation study gives more insights about our method.

Speech encryption algorithm using FFT and 3D-Lorenz–logistic chaotic map

In this paper a new speech encryption method using Fast Fourier Transform (FFT) and multiple chaotic maps has been developed for secured speech communication. In order to improve the drawbacks namely residual intelligibility in encrypted signal, poor quality in decrypted signal, low key space and high computational complexity that prevail in the exist speech encryption methods, a novel speech encryption algorithm based on a three dimension (3D) Lorenz-Logistic map has been developed. A new 3D Lorenz-Logistic map has been introduced by feeding Logistic map in 3D Lorenz map to obtain three different random number sequences. Totally eight initial and controlling parameters of 3D Lorenz-Logistic map are used as key values. The behavior of chaotic map is totally changed by changing the key values which generates highly randomized number sequence. Permutation and substitution plays vital role in this method. The input speech is applied to FFT to obtain real as well as imaginary values. Two random number sequences of the 3D Lorenz-Logistic map are used to permute the real and imaginary values of input speech signal. Remaining one random number sequence is used to permute a reference speech sample which is used for substitution of permuted real values of speech signal. The reverse process of permutation and substitution is used for recovering the desired signal. The Inverse Fast Fourier Transform (IFFT) is applied to reconstruct the original signal. The performance of the proposed method is verified using histogram analysis, spectrogram analysis, correlation analysis, signal to noise ratio analysis, NSCR (Number of Samples Changing Rate) and UACI (Unified Averaged Changed Intensity) analysis, key space and key sensitivity analysis, computational complexity measure, Perceptual Evaluation of Speech Quality (PESQ) analysis and subjective evaluation of speech quality. The results evidence that the proposed speech encryption method provides better security system with robust decryption quality.

FPGA implementation of a chaotic oscillator with odd/even symmetry and its application

We propose a mathematical system capable of exhibiting chaos with a chaotic attractor which is odd symmetrical in the x − y phase plane but even symmetrical in the x − z and y − z phase planes respectively. A hardware implementation of the system is done on a digital FPGA platform for verification. The system is also attractive in the sense that (i) its dynamics are single-parameter controlled and (ii) it inherently generates two chaotic clock signals. As an application, an FPGA design methodology using this oscillator for speech encryption is demonstrated. The security of the proposed encryption scheme is evaluated and results confirm its robustness. Due to the efficient hardware resource utilization, the encrypted system delivers a throughput of 1.3Gbit/sec using a Xilinx Kintex 7.

Perception of being observed by a speaker alters gaze behavior.

Previous research has shown that gaze behavior of a speaker's face during speech encoding is influenced by an array of factors relating to the quality of the speech signal and the encoding task. In these studies, participants were aware they were viewing pre-recorded stimuli of a speaker that is not representative of natural social interactions in which an interlocutor can observe one's gaze direction, potentially affecting fixation behavior due to communicative and social considerations. To assess the potential role of these factors during speech encoding, we compared fixation behavior during a speech-encoding task under two conditions: in the "real-time" condition, we used deception to convince participants that they were interacting with a live person who was able to see and hear them through online remote video communication. In the "pre-recorded" condition, participants were correctly informed they were watching a previously recorded video. We found that participants fixated the interlocutor's face significantly less in the real-time condition than the pre-recorded condition. When participants did look at the face, they fixated the mouth at a higher proportion of the time in the pre-recorded condition versus the real-time condition. These findings suggest that people engage in avoidance of potentially useful speech-directed fixations when they believe their fixations are being observed and demonstrate that social factors play a significant role in fixation behavior during speech encoding.

Speech Encryption Algorithm Based on Nonorthogonal Quantum State with Hyperchaotic Keystreams

With the advancement in modern computational technologies like cloud computing, there has been tremendous growth in the field of data processing and encryption technologies. In this contest there is an increasing demand for successful storage of the data in the encrypted domain to avoid the possibility of data breach in shared networks. In this paper, a novel approach for speech encryption algorithm based on quantum chaotic system is designed. In the proposed method, classical bits of the speech samples are initially encoded in nonorthogonal quantum state by the secret polarizing angle. In the quantum domain, encoded speech samples are subjected to bit-flip operation according to the Controlled–NOT gate followed by Hadamard transform. Complete superposition of the quantum state in both Hadamard and standard basis is achieved through Hadamard transform. Control bits for C-NOT gate as well as Hadamard gate are generated with a modified Lu˙-hyperchaotic system. Secret nonorthogonal rotation angles and initial conditions of the hyperchaotic system are the keys used to ensure the security of the proposed algorithm. The computational complexity of the proposed algorithm has been analysed both in quantum domain and classical domain. Numerical simulation carried out based on the above principle showed that the proposed speech encryption algorithm has wider keyspace, higher key sensitivity and robust against various differential and statistical cryptographic attacks.

A Classification Retrieval Method for Encrypted Speech Based on Deep Neural Network and Deep Hashing

In order to improve the retrieval efficiency and accuracy of the existing encrypted speech retrieval methods, and improve the semantic representation of speech features and classification performance, a classification retrieval method for encrypted speech based on deep neural network (DNN) and deep hashing is proposed. Firstly, the speech files are classified according to the category tags, and the speech files are encrypted by Rossler chaotic map method and uploaded to the cloud encrypted speech library. Secondly, the Log-Mel spectrogram features of the original speech are extracted, and extract deep semantic features and generate classification results through the trained convolutional neural network (CNN) and convolutional recurrent neural network (CRNN). Finally, the semantic feature hash code is obtained through the constructed hash function, combined with the category hash code encoded by One Hot coding to obtain the final deep hashing binary code, and uploaded to the deep hashing index table. When retrieval, the deep hashing binary code of the query speech is obtained, and the “two-stage” classification retrieval strategy and the normalized Hamming distance algorithm are used to match the semantic feature hash. Experimental results show that the proposed two DNN coding models have excellent feature learning performance, and has better recall rate, precision rate and retrieval efficiency.

SpEx: Multi-Scale Time Domain Speaker Extraction Network

Speaker extraction aims to mimic humans' selective auditory attention by extracting a target speaker's voice from a multi-talker environment. It is common to perform the extraction in frequency-domain, and reconstruct the time-domain signal from the extracted magnitude and estimated phase spectra. However, such an approach is adversely affected by the inherent difficulty of phase estimation. Inspired by Conv-TasNet, we propose a time-domain speaker extraction network (SpEx) that converts the mixture speech into multi-scale embedding coefficients instead of decomposing the speech signal into magnitude and phase spectra. In this way, we avoid phase estimation. The SpEx network consists of four network components, namely speaker encoder, speech encoder, speaker extractor, and speech decoder. Specifically, the speech encoder converts the mixture speech into multi-scale embedding coefficients, the speaker encoder learns to represent the target speaker with a speaker embedding. The speaker extractor takes the multi-scale embedding coefficients and target speaker embedding as input and estimates a receptive mask. Finally, the speech decoder reconstructs the target speaker's speech from the masked embedding coefficients. We also propose a multi-task learning framework and a multi-scale embedding implementation. Experimental results show that the proposed SpEx achieves 37.3%, 37.7% and 15.0% relative improvements over the best baseline in terms of signal-to-distortion ratio (SDR), scale-invariant SDR (SI-SDR), and perceptual evaluation of speech quality (PESQ) under an open evaluation condition.

An Encrypted Speech Retrieval Method Based on Deep Perceptual Hashing and CNN-BiLSTM

Since convolutional neural network (CNN) can only extract local features, and long short-term memory (LSTM) neural network model has a large number of learning calculations, a long processing time and an obvious degree of information loss as the length of speech increases. Utilizing the characteristics of autonomous feature extraction in deep learning, CNN and bidirectional long short-term memory (BiLSTM) network are combined to present an encrypted speech retrieval method based on deep perceptual hashing and CNN-BiLSTM. Firstly, the proposed method extracts the Log-Mel Spectrogram/MFCC features of the original speech and enters the CNN and BiLSTM networks in turn for model training. Secondly, we use the trained fusion network model to learn the deep perceptual feature and generate deep perceptual hashing sequences. Finally, the normalized Hamming distance algorithm is used for matching retrieval. In order to protect the speech security in the cloud, a speech encryption algorithm based on a 4D hyperchaotic system is proposed. The experimental results show that the proposed method has good discrimination, robustness, recall and precision compared with the existing methods, and it has good retrieval efficiency and retrieval accuracy for longer speech. Meanwhile, the proposed speech encryption algorithm has a high key space to resist exhaustive attacks.

Robust Speech Filter and Voice Encoder Parameter Estimation Using the Phase–Phase Correlator

In recent years, linear prediction voice encoders have become very efficient in terms of computing execution time and channel bandwidth usage while providing, in the absence of impulsive noise, natural sounding synthetic speech signals. This good performance has been achieved via the use of a maximum likelihood parameter estimation of an auto-regressive model of order ten that best fits the speech signal under the assumption that the signal and the noise are Gaussian stochastic processes. However, this method breaks down in the presence of impulse noise, which is common in practice, resulting in harsh or non-intelligible audio signals. In this paper, we propose a robust estimator of correlation, the Phase-Phase correlator that is able to cope with impulsive noise. Utilizing this correlator, we develop a Robust Mixed Excitation Linear Prediction encoder that provides improved audio quality for voiced, unvoiced, and transition speech segments. This is achieved by applying a statistical test to robust Mahalanobis distances for identifying the outliers in the corrupted speech signal, which are then replaced with filtered signals. Simulation results reveal that the proposed estimator of correlator outperforms in variance, bias, and breakdown point compared to three other robust approaches based on the arcsin law, the polarity coincidence correlator, and the median-of-ratio estimator without sacrificing the encoder bandwidth efficiency and the compression gain while remaining compatible with real-time applications. Furthermore, in the presence of impulsive noise, the proposed speech encoder speech subjective quality outperforms the state-of-the-art in terms of mean opinion score.

An encrypted speech retrieval algorithm based on Chirp-Z transform and perceptual hashing second feature extraction

In order to satisfy the requirements of retrieval time-efficiency and security for encrypted speech data retrieval in the cloud environment, and to improve the impact of noise on the robustness and discrimination for the speech perceptual hashing scheme, an encrypted speech retrieval algorithm based on Chirp-Z transform and perceptual hashing second feature extraction is proposed in this paper. The speech owner first processes the original speech file by pre-processing, framing, and adding window. The features of the original speech file is extracted by Chirp-Z transform combined with the sparse random matrix to construct a hash sequence. Then encrypt the original speech file based on the m sequence and upload it to the cloud to ensure the security of information. By processing the speech perceptual hashing feature, the speech features are re-extracted, and the speech is evenly classified by k-means clustering technique. The binary string of several hundred bits is converted into a decimal number. Finally, the second feature is stored in system hash index table of the cloud. When the user retrieves, the query speech is denoised and the hash sequence is extracted. Then the secondary features of the hash sequence are extracted and matched with the encrypted speech features in the cloud system hash index table to obtain the retrieval result. The experimental results show that the proposed algorithm greatly compresses the information capacity of speech features, significantly improves the retrieval time-efficiency, with strong robustness and discrimination, and has a good retrieval effect on noisy speech.

Behaviour, biology and evolution of vocal learning in bats.

The comparative approach can provide insight into the evolution of human speech, language and social communication by studying relevant traits in animal systems. Bats are emerging as a model system with great potential to shed light on these processes given their learned vocalizations, close social interactions, and mammalian brains and physiology. A recent framework outlined the multiple levels of investigation needed to understand vocal learning across a broad range of non-human species, including cetaceans, pinnipeds, elephants, birds and bats. Here, we apply this framework to the current state-of-the-art in bat research. This encompasses our understanding of the abilities bats have displayed for vocal learning, what is known about the timing and social structure needed for such learning, and current knowledge about the prevalence of the trait across the order. It also addresses the biology (vocal tract morphology, neurobiology and genetics) and evolution of this trait. We conclude by highlighting some key questions that should be answered to advance our understanding of the biological encoding and evolution of speech and spoken communication. This article is part of the theme issue 'What can animal communication teach us about human language?'

Speech Encryption Based on Zaslavsky Map

Speech Encryption Based on Zaslavsky Map

Effect of congenital toxoplasmosis on the encoding of speech in infants

ObjectiveTo investigate the effect of congenital toxoplasmosis (CTP) on the Frequency-Following Response (FFR) in infants. Study design11 infants diagnosed with CTP and 12 healthy infants with no risk indicators for hearing impairment, aged 29–90 days old. All infants underwent an FFR neurophysiological assessment. The test stimulus was the syllable [da], 40 ms in duration, which was monaurally presented to the right ear at an intensity of 80 dBnHL. Absolute latencies and amplitudes of the V, A, C, D, E, F, and O waves, the slope (μV/ms) and measure between onset (A) and offset (O), were compared between the two groups. ResultsInfants with CTP had increased latency of FFR waves V, A, E, F, and O, and decreased amplitude for waves A and F. They also showed a reduction in A–O slope and a higher latency difference between onset (A) and offset (O). ConclusionThe neurophysiological responses of Frequency-Following Response can be influenced by congenital toxoplasmosis. Since, the CTP showed prolongation of the V, A, E, F and O waves and decrease of the amplitude for waves A and F.

Speech Encryption using Advanced Encryption Standard for Secured Communication

Securing speech communication is a great challenging task and interesting research area in recent years. Transmission of voice data over untrusted networks leads to attack the confidential information. Various speech encryption algorithms are in progress and even though speech data have been encrypted some additional features are still getting compromised to the eavesdroppers which are a get setback for security attacks. The state-of-art of Advanced Encryption Standard (AES) algorithm when compared to other encryption algorithm it has different types of cipher key length such as 128,192 and 256 bits respectively. Advanced Encryption Standard is analyzed statistically for revealing its superiority based on confusion and diffusion properties for evaluating the robustness against statistical attacks. A novel approach is developed for encrypting speech data using Advanced Encryption Standard (AES). Encrypted and the decrypted signal are evaluated based on some quality analysis for immunity checking against frequency-domain attack, brute force attack and statistical attack etc.

Encoding of speech in noise in adults using hearing aids: effect of noise reduction algorithm

Introduction: Subjects using hearing aids complains of difficulty in speech understanding in noise. Speech auditory brainstem response (S-ABR) provides cues for temporal and spectral encoding of speech in the brainstem.Objective: This work was designed to evaluate speech processing in patients fitted with hearing aid using S-ABR and to study the effect of noise reduction algorithm on speech processing in patients fitted with hearing aids.Materials and method: This study included 20 adults with mild to severe SNHL fitted with HA at least 3 months with aided threshold in all frequencies better than or equal to 30 dB. Hearing aid was used with all the subjects and was programed for each subject according to his audiogram followed by aided evaluation using S-ABR. Through sound field in four different test conditions with NR algorithm on in quiet and in the presence of noise then with the NR off in quiet and in the presence of noise were performed.Results: There was a statistically significant delay in the S-ABR onset and offset in noise compared to quiet. The effect of noise on S-ABR while the NR was on showed statistically significant difference in the onset response in both quiet and noise. When the NR was off, there was marked effect of noise on both the onset response and offset response.Conclusions: It was found that hearing aid users are vulnerable to noise regarding the speech encoding at the level of the brainstem. Noise reduction algorithm maintains the neural activity in response to speech and improves speech perception in noise.

Masked Speech Recognition in School-Age Children

Children who are typically developing often struggle to hear and understand speech in the presence of competing background sounds, particularly when the background sounds are also speech. For example, in many cases, young school-age children require an additional 5- to 10-dB signal-to-noise ratio relative to adults to achieve the same word or sentence recognition performance in the presence of two streams of competing speech. Moreover, adult-like performance is not observed until adolescence. Despite ample converging evidence that children are more susceptible to auditory masking than adults, the field lacks a comprehensive model that accounts for the development of masked speech recognition. This review provides a synthesis of the literature on the typical development of masked speech recognition. Age-related changes in the ability to recognize phonemes, words, or sentences in the presence of competing background sounds will be discussed by considering (1) how masking sounds influence the sensory encoding of target speech; (2) differences in the time course of development for speech-in-noise versus speech-in-speech recognition; and (3) the central auditory and cognitive processes required to separate and attend to target speech when multiple people are speaking at the same time.

FPGA Design of Encryption Speech system using Synchronized Fixed-Point Chaotic Maps Based Stream Ciphers

In this work, speech encryption using synchronized fixed-point chaotic map-based stream ciphers (SFPCM-SC) is suggested. Five chaotic maps named quadratic, henon, logistic, lozi and duffing are synchronized by using master-slave synchronization technique and then used to generate the pseudo random bit generator (PRBG) using fixed-point converter. The PRBG is then Xor-ed with digitized speech signal where the encrypted signal is created. In the other side, the same map is synchronized with the master one and used to recover the original speech signal. First this work is simulated by using MATLAB and then built the design using Xilinx system generator (XSG). Finally, the hardware co-simulation is applied for the proposed system by using FPGA SP605 XC6SLX45T device. The results show that the error between master and slave become zero after a small period and the original speech signal is recovered with real time environment in successful.