Spectrum Of Speech Signal Research Articles

IT Diagnostics of Parkinson’s Disease Based on the Analysis of Voice Markers and Machine Learning

The results of studying the parameters of the spectra of speech signals by machine learning with the use of neural networks are presented. This study was carried out in order to confirm experimentally the possibility of performing an assessment of these parameters for the detection of Parkinson’s disease in the early stages (IT diagnostics). During the study, the public database was used, which systematized the spectra of vowel sounds uttered by patients with Parkinson’s disease. The applied method is binary data classification. In the course of the study, the speech data spectrum was first preprocessed, which consisted of filtering it in order to remove its noise components and eliminate bursts and gaps in it. Then the parameters of the processed spectrum of speech data were determined: average value, maximum and minimum, peak, wavelet coefficients, MFCC and TQWT. After that, the object was selected using the PCA algorithm. The model was trained using the Knn and Random Forest algorithms, as well as the Bayesian neural network. The Bayesian optimization algorithm and the GridSearch method were used to find the best model hyperparameters. It has been established that when using Knn, Random Forest and Bayesian neural network, it is possible to increase the accuracy of recognition of Parkinson’s disease by 94.7; 88.16 and 74.74 %, respectively. A similar study by other scientists showed that the recognition accuracy of data sets was only 86 %.

Packet Loss Concealment Based on Phase Correction and Deep Neural Network

In a packet switching network, the performance of packet loss concealment (PLC) is often affected by inaccurate estimation of phase spectrum of speech signal in the lost packet. In order to solve this problem, two kinds of PLC methods in the scene of continuous packet loss are proposed in this paper based on phase correction. One of them is based on waveform similarity overlap-add (WSOLA) and deep neural network (DNN), and the other one is based on the Griffin–Lim algorithm (GLA) and DNN. In the first method, considering the correlation of adjacent frames of speech signal, the periodicity of speech signal is well retained by the WSOLA method so that the phase spectrum of the lost signal is recovered. Combined with the prediction of amplitude spectrum of the lost signal by the DNN, the performance of the PLC in the case of continuous packet loss is effectively improved. In the second method, the phase spectrum of the lost signal is modified iteratively by the amplitude spectra estimated via the DNN based on the consistency of Fourier transform so that the phase spectra match the amplitude spectra. The PLC is greatly achieved as well. The experimental results show that the proposed PLC methods provide better speech quality than the reference methods.

Improving the potential of Enhanced Teager Energy Cepstral Coefficients (ETECC) for replay attack detection

In the scope of voice biometrics, the term replay attack, (RA) refers to the dishonest attempt made by an impostor to spoof someone else’s identity by replaying the subject’s previously recorded speech close to the Automatic Speaker Verification (ASV) system under attack. State-of-the-art strategies for RA detection, such as the Enhanced Teager Energy Cepstral Coefficients (ETECC), have shown promising results due to their precision in measuring energy from high frequency components of speech, as a function of two recently defined concepts: signal mass and Enhanced Teager Energy Operator (ETEO). Nevertheless, since the replay mechanism prominently deteriorates the speech signal spectrum just in those spectral zones, we propose the association of ETEO with different strategies to further improve the previous results in getting effective countermeasures against RAs. Specifically, comprehensive evaluations which include a detailed mathematical analysis, a simulation on amplitude and frequency modulated (AM–FM) signals, and a spectrographic inspection involving different filterbank structures, along with their experimental results, are provided in this paper. In addition, ETEO-derived features are contrasted to existing feature sets by using Paraconsistent Feature Engineering (PFE) for feature ranking, expanding our previously published results. Lastly, experiments are performed with ASVSpoof-2017 version 2.0 dataset, Realistic Replay Attack Microphone Array Speech Corpus (ReMASC), BTAS-2016, dataset, ASVSpoof-2019 challenge dataset, and ASVSpoof-2015 challenge dataset, considering Gaussian Mixture Models (GMMs), Convolutional Neural Networks (CNNs) and Light-CNN architectures as being the classifiers. The standalone ETECC-GMM system showed the best performance by producing equal error rates (EERs) of 5.55% and 10.75% on development and evaluation sets, respectively.

Combination of sparsity of direct waves and detection of signal for distant speech enhancement

A joint dereverberation and denoising framework is proposed for enhancement of distant speech in a conference room. The sparsity of speech multipath propagation is exploited by a microphone array for suppressing late reverberation. The dereverberation speech is transformed to solve a minimization problem where the sparse prediction coefficients are estimated by a least-square iteration algorithm. The residual noise components are removed by iteratively reconstructing the time–frequency spectrum of speech signals. Detection of speech presence is carried out over each cell of the spectrum based on the Bayes rule. The data analysis and the real-time processing results have shown that the proposed joint framework outperforms the existing approaches in terms of objective metrics of speech quality.

An improved framework for Parkinson’s disease prediction using Variational Mode Decomposition-Hilbert spectrum of speech signal

Parkinson’s disease (PD) is a neuro-degenerative disease due to loss of brain cells, which produces dopamine. It is most common after Alzheimer’s disease specially seen in old age people. In the earlier stage of disease, it has been noticed that most of the people suffering from speech disorder. From last two decades many studies have been conducted for the analysis of vocal tremors in PD. This study explores the combined approach of Variational Mode Decomposition (VMD) and Hilbert spectrum analysis (HSA) to investigate the voice tremor of patients with PD. A new set of features Hilbert cepstral coefficients (HCCs) are proposed in this study. Proposed features are assessed using vowels and words of PC-GITA database. The effectiveness of HCC features is utilized to perform classification, and regression analysis for PD detection. The highest average classification accuracy up to 91% and 96% is obtained with vowel /a/ and word /apto/ respectively. Further the classification accuracy up to 82% is obtained with independent dataset, when tested with the optimized model developed using PC-GITA database. In dysarthria level prediction highest correlation up to 0.82 is obtained using vowel /a/ and 0.8 with word /petaka/. The outcomes of this study indicate that the proposed articulatory features are suitable and accurate for PD assessment.

An Iterative Graph Spectral Subtraction Method for Speech Enhancement

In this paper, we investigate the application of graph signal processing (GSP) theory in speech enhancement. We first propose a set of shift operators to construct graph speech signals, and then analyze their spectrum in the graph Fourier domain. By leveraging the differences between the spectrum of graph speech and graph noise signals, we further propose the graph spectral subtraction (GSS) method to suppress the noise interference in noisy speech. Moreover, based on GSS, we propose the iterative graph spectral subtraction (IGSS) method to further improve the speech enhancement performance. Our experimental results show that the proposed operators are suitable for graph speech signals, and the proposed methods outperform the traditional basic spectral subtraction (BSS) method and iterative basic spectral subtraction (IBSS) method in terms of both signal-to-noise ratios (SNR) and mean Perceptual Evaluation of Speech Quality (PESQ).

ЭФФЕКТИВНОСТЬ ПЕРЕДАЧИ ИНФОРМАЦИИ СИСТЕМ ОПОВЕЩЕНИЯ И ТЕЛЕКОММУНИКАЦИЙ АУДИООБМЕНА В УСЛОВИЯХ ПОМЕХ

In the paper shown there is presented an approach to the solution of the problem of the effectiveness parameter assessment in telecommunication systems of operational and command communication, systems of warning speakerphone and audio-exchange. 
 There are considered the matters of the dependence investigation of the acoustic speech signal/noise ratio to the assurance of the required syllabic legibility for the possibility to increase the function effectiveness of telecommunication systems and information exchange operated under complex noise situation. There is shown the dependence of formant legibility upon the meaning of average geometric frequencies in each i-th band of a frequency spectrum of acoustic speech signals. The degree of the impact upon syllabic legibility of the acoustic speech signal/noise ratio is shown. In the paper it is shown that for obtaining speech information with syllabic legibility higher than 93% required for complete perception by a subscriber it is necessary to ensure the acoustic signal/noise ratio at the level no less than 20 dB. The problems in the probability density approximation of acoustic signals with the use of generalized polynomials on function basis systems are presented.

Speaker Verification Using Autoregressive Spectrum of Speech Signal in Composite Vector Stochastic Processes Model Representation

Speaker Verification Using Autoregressive Spectrum of Speech Signal in Composite Vector Stochastic Processes Model Representation

Analysis of Spectral Characteristics of Ukrainian and Several European Languages

A number of scientific, technical, medical and linguistic tasks require the establishment of an acoustic affinity or a peculiarity of the Ukrainian language in comparison with European languages. Today these issues are not thoroughly investigated.The objective parameters of the speech process as a random signal are not given sufficient attention. And understanding the processes of how the brain receives information can be obtained by analyzing the relationship of the acoustic characteristics of the signal with its phonetic parameters.The most objective acoustic characteristic of a language is its integral characteristic, namely the averaged spectral density of the speech signal power. The first assessment of this characteristic for the Ukrainian language was made by the authors in previous work. The purpose of the article is to study the averaged spectra of languages of the Indo-European family, namely, Russian, Polish, Italian and English, as well as to compare them with the corresponding spectral characteristics of the Ukrainian language.Audiobook audio recordings from the electronic resources indicated in the text were used for the research. The analysis was performed for two styles of language: artistic and nonfiction, followed by averaging the results.To find the average power density spectrum of speech signals, their processing was performed in the MATLAB® software environment.The duration of the studied records was 2 minutes. 30 s. ± 5 s. The tested frequency range is selected from 100 Hz to 20 kHz.The results obtained are presented in the form of graphical representations of the spectral characteristics of linguistic signals averaged over styles in comparison with the Ukrainian language.The basic regularities of the spectrum of the Ukrainian language are analyzed. The high level of spectral components of the low-frequency range is characteristic, which indicates the thoracic nature of the loud sounds. Some decrease in the level of the spectrum of the Ukrainian language in the midrange indicates a lower representation of ringing consonants and, in general, no emphasizing of consonant sounds. The high frequency formant with a maximum in the region of 4 kHz, which can be called a "singing formant", is essential. At higher frequencies there is a significant decrease in the level of the spectrum, which indicates the absence of whistles and whispers in the Ukrainian language. The results obtained are in line with the main conclusions of linguists who distinguish the particular sweetness of the Ukrainian language compared to RussianA comparison of the spectra of Ukrainian and Polish obvious that they are sufficiently similar to the frequency of about 1.5 kHz. At higher frequencies in Polish the presence of clanging and rustling sounds, while in the Ukrainian language appears "singing formant" and there is a sharp decline in the levels of spectral components.The level of components of the spectrum of the Italian language is much lower than in the Ukrainian, in the frequency range 100-1500 Hz. Italian has powerful high-frequency spectral components, with its high-frequency formants also occur in the region of 4 kHz, but a more significant rise in frequency response is observed within 7-9 kHz.There is a significant similarity between the English and Ukrainian spectra at almost the frequency of 1800 Hz. Further up to 8 kHz, while the level of high-frequency components is maintained in English, a “singing” formant is formed in the Ukrainian language. After 8 kHz, the spectra fall almost equally, unlike the other languages considered.

Wavelet-Based Power Normalized Spectrum for Hindi Phoneme Classification

This paper presents wavelet-based power normalized spectrum for computing robust cepstral features named WP-PNCC features. The proposed technique computes wavelet packet-based short-time spectrum of speech signal. A nonlinear function is defined as relating power spectrum of clean speech to the power spectrum of speech corrupted with noise. The constants of function are computed from longer-duration speech spectrum, and the short-time spectrum for each frame is weighted with the power function. The weighted speech spectrum is processed with logarithmic and discrete cosine transform operation to compute cepstral coefficients. The cepstral coefficients thus obtained are processed with quantile-based cepstral dynamics normalization technique. The proposed features are examined with hidden Markov model classifier on TIFR database for Hindi phoneme classification task and on TIMIT database for English phoneme classification task along with mel-frequency cepstral coefficients, power normalized cepstral coefficients and 24-band wavelet-based features in clean and noisy environments. Different noises from NOISEX-92 database are used for preparing noisy database with SNR ranging from 20 dB to 0 dB. The results show enhanced performance of proposed features in all the considered cases. The simulations are performed on MATLAB 2015b. The performance of proposed features is also evaluated on hidden Markov model toolkit-based speech recognition system. The comparative results confirm the robustness of proposed features with sufficient improvement over other features examined in this paper.

Vocal Source Contribution to Speaker Recognition

The vocal source and the pulse shape of the glottal flow are determined through the regularized ratio of the speech signal spectra at the intervals of the open and closed vocal slit within each period of the fundamental tone. Three databases were used: Russian numerals for 216 men and 177 women, the base obtained by converting the Russian database by the codec on 9.2 kbps, and the TIMIT database. The pitch period and 7 coefficients for the principal components of the glottal flow provide an average error of recognizing males below 8% for a sequence of 6 vowels. The minimum average recognition error for the initial base of Russian numerals for females makes about 15%, for males in the codec database makes about 15%, and for males in the TIMIT makes about 44%. The minimum average error of males’ recognition in the space of 7 coefficients for the principal components in the Russian database makes about 26%, but about 27% of the speakers have an average error of less than 10%.

Automatic Croup Diagnosis Using Cough Sound Recognition.

Croup, a respiratory tract infection common in children, causes an inflammation of the upper airway restricting normal breathing and producing cough sounds typically described as seallike "barking cough." Physicians use the existence of barking cough as the defining characteristic of croup. This paper aims to develop automated cough sound analysis methods to objectively diagnose croup. In automating croup diagnosis, we propose the use of mathematical features inspired by the human auditory system. In particular, we utilize the cochleagram for feature extraction, a time-frequency representation where the frequency components are based on the frequency selectivity property of the human cochlea. Speech and cough share some similarities in the generation process and physiological wetware used. As such, we also propose the use of mel-frequency cepstral coefficients which has been shown to capture the relevant aspects of the short-term power spectrum of speech signals. Feature combination and backward sequential feature selection are also experimented with. Experimentation is performed on cough sound recordings from patients presenting various clinically diagnosed respiratory tract infections divided into croup and non-croup. The dataset is divided into training and test sets of 364 and 115 patients, respectively, with automatically segmented cough sound segments. Croup and non-croup patient classification on the test dataset with the proposed methods achieve a sensitivity and specificity of 92.31% and 85.29%, respectively. Experimental results show the significant improvement in automatic croup diagnosis against earlier methods. This paper has the potential to automate croup diagnosis based solely on cough sound analysis.

Auditory distraction by speech: Can a babble masker restore working memory performance and subjective perception to baseline?

Background speech has a detrimental impact on employees’ privacy perception and cognitive performance in open-plan offices. Sound masking that covers speech sounds can help to improve the speech privacy and the ability to work undisturbed in open office environments. Recently, non-artificial sounds such as babble or water sounds have been suggested as masking sounds because they may be perceived as more natural and subsequently less annoying. This paper compares the working memory performance and annoyance perception during background speech that was masked by different babble sounds and a waterfall sound to speech that was masked by stationary noise with the frequency spectrum of the distracting background speech signal (target speech). In a first laboratory experiment the effects of different babble sounds were compared with the noise sound at a speech-to-noise ratio of −3 dB. All subjects had to complete a serial short-term memory task and a questionnaire that covered subjective ratings. All sound conditions with masked target speech produced similar error rates and annoyance ratings as unmasked target speech. In a second laboratory experiment, the speech-to-noise ratio of target speech that was masked by babble consisting of 48 target speech signals was varied between −6 and −12 dB in steps of 3 dB. Target based babble resulted in similar error rates as stationary noise at −6 dB speech-to-noise ratio but the findings suggest that target speech that is masked by babble might even be perceived as more annoying than speech that is masked by stationary target spectrum based noise.

Incorporating GammaTone filterbank and Welch spectral estimation in subband processing-based localization of multiple simultaneous speakers

This research addresses the problem of multiple speakers localization by applying Generalized Cross-Correlation (GCC)-band techniques on different frequency subbands. To improve the localization performance in the presence of noise and reverberation, we promote the baseline method in two folds; in the first part, human auditory-based filters are proposed for subband processing. GammaTone filterbank is the most important group of these filters. The GammaTone filterbank increases the frequency resolution in low frequencies and provides better utilization of the speech spectral information. The GammaTone filter is applied to the whole speech signal spectrum. In the second part of the proposed method, the aim is to consider the speech spectra more intelligently in the process of the localization. To this end, we use different spectral estimation methods to select the part of speech spectrum that has useful information. By comparing various spectral estimation techniques, Welch method has been selected as the best one for speech spectrum estimation. We can decrease the computational complexity and increase the accuracy using Welch estimation method. The proposed method has been examined in different acoustical environments for real and simulated data to show its superiority over the other methods.

Selecting frames for automatic speech recognition based on acoustic landmarks

Most mainstream Mel-frequency cepstral coefficient (MFCC) based Automatic Speech Recognition (ASR) systems consider all feature frames equally important. However, the acoustic landmark theory disagrees with this idea. Acoustic landmark theory exploits the quantal non-linear articulatory-acoustic relationships from human speech perception experiments and provides a theoretical basis of extracting acoustic features in the vicinity of landmark regions where an abrupt change occurs in the spectrum of speech signals. In this work, we conducted experiments, using the TIMIT corpus, on both GMM and DNN based ASR systems and found that frames containing landmarks are more informative than others during the recognition process. We proved that altering the level of emphasis on landmark and non-landmark frames, through re-weighting or removing frame acoustic likelihoods accordingly, can change the phone error rate (PER) of the ASR system in a way dramatically different from making similar changes to random frames. Furthermore, by leveraging the landmark as a heuristic, one of our hybrid DNN frame dropping strategies achieved a PER increment of 0.44% when only scoring less than half, 41.2% to be precise, of the frames. This hybrid strategy out-performs other non-heuristic-based methods and demonstrated the potential of landmarks for computational reduction for ASR.

Speech Bandwidth Extension Aided by Magnitude Spectrum Data Hiding

Public telephone systems transmit speech across a limited frequency range, about 300–3400 Hz, called narrowband (NB) which results in a significant reduction of quality and intelligibility of speech. This paper proposes a fully backward compatible novel method for bandwidth extension of NB speech. The method uses magnitude spectrum data hiding technique to provide a perceptually better wideband speech signal. The spectral envelope parameters are extracted from the down-sampled frequency shifted version of the high-frequency components of speech signal existing above NB, which are then encoded and spread by using spreading sequences, and are embedded in the low-amplitude high-frequency regions of the magnitude spectrum of NB speech signal. The embedded information is extracted at the receiving end to reconstruct the wideband speech signal. Theoretical and simulation analyses show that the proposed method is robust to quantization and channel noises. The comparison category rating listening and log spectral distortion tests clearly show that the reconstructed wideband signal gives a much better performance in terms of speech quality when compared to the conventional speech bandwidth extension methods employing data hiding.

Bandwidth extension of telephone speech using magnitude spectrum data hiding

Public telephone systems transmit speech across a limited frequency range, about 300---3400 Hz, called narrowband (NB) which results in a significant reduction of quality and intelligibility of speech. This paper proposes a fully backward compatible novel method for bandwidth extension of NB speech. The method uses magnitude spectrum data hiding technique to provide a perceptually better wideband speech signal. Code excited linear prediction parameters are extracted from the down sampled frequency shifted version of the high frequency components of speech signal existing above NB, which are spread by using pseudo-noise codes, and are embedded in the low amplitude high-frequency regions of the magnitude spectrum of NB speech signal. The embedded information is extracted at the receiving end to reconstruct the wideband speech signal. Theoretical and simulation analyses show that the proposed method is robust to quantization and channel noises. The comparison category rating listening and log spectral distortion tests clearly show that the reconstructed wideband signal gives a much better performance in terms of speech quality when compared to some of the existing speech bandwidth extension methods employing data hiding.

Determination of a vocal source by the spectral ratio method

The inverse problem with respect to functions proportional to a voice source and volume velocity of the air flow through the glottis is solved as follows: we compute the inverse Fourier transform of the regularized fraction of short-term speech-signal spectra at intervals with an opened (closed) glottis, minimizing the optimality criterion with respect to the regularization parameter and the glottis opening (closing) time. The optimality criterion for solutions includes the values of the volume velocity and its time derivative at the ends of the interval with an opened glottis and the total value of the negative volume velocity. To obtain an empirical error estimate for the solution, experiments using synthesized signals with various parameters, direct measurements of the glottis, and signals synchronously recorded through pairs of microphones of different types are performed. The most probable determination error for the volume velocity is less than 5% for synthetic sources; if the area of the glottis of the source is measured experimentally, then the said error is about 10%. The discrepancy of solutions for the same signal synchronously recorded through a pair of microphones of different types is less than 10%.

Classification of speech under stress using harmonic peak to energy ratio

This paper explores the analysis and classification of speech under stress using a new feature, harmonic peak to energy ratio (HPER). The HPER feature is computed from the Fourier spectra of speech signal. The harmonic amplitudes are closely related to breathiness levels of speech. These breathiness levels may be different for different stress conditions. The statistical analysis shows that the proposed HPER feature is useful in characterization of various stress classes. Support Vector Machine (SVM) classifier with binary cascade strategy is used to evaluate the performance of the HPER feature using simulated stressed speech database (SSD). The performance results show that the HPER feature successfully characterizes different stress conditions. The performance of the HPER feature is compared with the mel frequency cepstral coefficients (MFCC), the Linear prediction coefficients (LPC) and the Teager-Energy-Operator (TEO) based Critical Band TEO Autocorrelation Envelope (TEO-CB-Auto-Env) features. The proposed HPER feature outperforms the MFCC, LPC and TEO-CB-Auto-Env features. The combination of the HPER feature with the MFCC feature further increases the system performance.

A novel method for early diagnosis of Alzheimer's disease based on higher-order spectral estimation of spontaneous speech signals.

One main challenge for medical investigators is the early diagnosis of Alzheimer's disease (AD) because it provides greater opportunities for patients to be eligible for more clinical trials. In this study, higher order spectra of human speech signals during AD were explored to analyze and compare the quadratic phase coupling of spontaneous speech signals for healthy and AD subjects using bispectrum and bicoherence. The results showed that the quadratic phase couplings of spontaneous speech signal of persons with Alzheimer's were reduced compared to healthy subject. However, the speech phase coupled harmonics shifted to the higher frequencies in Alzheimer's than healthy subjects. In addition, it was shown not only are there significant differences between Alzheimer's and control subjects in parameters estimated, but also the speech patterns appeared to have fluctuated in both types of spontaneous speech.