Voice Classification Research Articles

A Two-Stage Cepstral Analysis Procedure for the Classification of Rough Voices

The objective of this study was to investigate the ability of a two-stage method of cepstral peak identification to effectively discriminate rough vs breathy vs typical voice in sustained vowel productions. It was hypothesized that a dual-stage search for cepstral peak prominences (CPP's) above and below specified quefrency/F0 cutoffs would result in a CPP difference that would be characteristic of the rough, diplophonic voice type. Central one-second portions of sustained vowel /a/ productions were obtained from 90 subjects (rough, breathy, and normophonic voices). All voice samples were analyzed using a a two-stage cepstral analysis process in which a CPPHigh-Low difference value was obtained by identifying cepstral peaks above and below a lower limit for expected F0 (150Hz for females and 90Hz for males), called CPPHigh and CPPLow respectively. The CPPHigh-Low difference value was observed to be a highly significant predictor, with negative values for this parameter characteristic of a dominant subharmonic in the voice signal and the perception of diplophonic, rough voice. Correct classification of rough vs nonrough voice samples was 82.2% (sensitivity 0.80 and specificity 0.833). In the consideration of three group classification (breathy vs. normophonic vs. rough), models incorporating two predictors (the CPP obtained from a single search through a 60 to 300Hz frequency range (CPPDefault) and the CPPHigh-Low difference value) correctly classified 78.88% of the voice samples. Rough, diplophonic voices were consistently observed to have a subharmonic peak that was greater in amplitude than the cepstral peak obtained within the region of the expected F0, resulting in a negative value for the CPPHigh-Low difference. The two-stage cepstral analysis process described herein is visually intuitive from the graphical display of a cepstrum and is a simple extended calculation derived from cepstral analysis procedures that have been recommended as essential in the acoustic description of vocal quality.

Applied Chaos Level Test for Validation of Signal Conditions Underlying Optimal Performance of Voice Classification Methods.

The purpose of this study is to introduce a chaos level test to evaluate linear and nonlinear voice type classification method performances under varying signal chaos conditions without subjective impression. Voice signals were constructed with differing degrees of noise to model signal chaos. Within each noise power, 100 Monte Carlo experiments were applied to analyze the output of jitter, shimmer, correlation dimension, and spectrum convergence ratio. The computational output of the 4 classifiers was then plotted against signal chaos level to investigate the performance of these acoustic analysis methods under varying degrees of signal chaos. A diffusive behavior detection-based chaos level test was used to investigate the performances of different voice classification methods. Voice signals were constructed by varying the signal-to-noise ratio to establish differing signal chaos conditions. Chaos level increased sigmoidally with increasing noise power. Jitter and shimmer performed optimally when the chaos level was less than or equal to 0.01, whereas correlation dimension was capable of analyzing signals with chaos levels of less than or equal to 0.0179. Spectrum convergence ratio demonstrated proficiency in analyzing voice signals with all chaos levels investigated in this study. The results of this study corroborate the performance relationships observed in previous studies and, therefore, demonstrate the validity of the validation test method. The presented chaos level validation test could be broadly utilized to evaluate acoustic analysis methods and establish the most appropriate methodology for objective voice analysis in clinical practice.

CT-based Morphometric Analysis of Professional Opera Singers' Vocal Folds

Our work aimed to study the relationship between the length of vocal folds and classical voice type among professional opera singers. Also, the present paper attempts to assess if individual features such as height and body mass index are associated with the vocal folds length (VFL) and the vocal range. Our research compared VFL and voice type by evaluation of neck computed tomography imaging of professional opera singers obtained between 2009 and 2015. Vocal range was measured after scanning procedure by a speech-language pathologist. The results were used to compare VFL with vocal range, voice type, and individual features of every singer. The group included 93 opera singers (female: 31 sopranos, 15 mezzos, and 5 contraltos; male: 17 tenors, 16 baritones, 6 basses, and 3 countertenors). Different values such as VFL, vocal range, and individuals' features (height, weight, body mass index) were analyzed statistically. Some significant differences and correlation were found. The VFL was diversified between classical voice types. VFL for soprano, mezzo-soprano, and bass was significantly different from every other type of voice. The vocal range has been shown to have a negative correlation with VFL value, which reflects the narrowest vocal range among bass voices. The analysis confirmed a linear correlation between VFL and individual features of the body such as height and body mass index. Presented data support the hypothesis that there are significant differences of VFL, vocal range, and body morphology between professional singers of different voice classification.

A methodology for voice classification based on the personalized fundamental frequency estimation

Abstract Nowadays, the incidence of voice disorders is increasing rapidly, with about a third of the population suffering from dysphonia at some point in their lives. Dysphonia is a disorder that alters vocal quality and can impair and reduce the quality of life. The structural or functional alteration of the phonatory apparatus, unhealthy lifestyles or an excessive use of the vocal cords for work activities (e.g. teaching) can cause voice disorders. Unfortunately, people who suffer from dysphonia often underestimate its symptoms and therefore delay consulting a speech therapist for accurate voice assessment and treatment. Voice disorder evaluation involves a series of tests, including an acoustic analysis. This quantifies the measurements of voice quality through the evaluation of certain characteristic parameters, for example the fundamental frequency (F0). In this paper, a personalized methodology for the estimation of the F0 is presented. The personalization is accomplished by taking into account two of the main factors that influence the F0, the gender and age of the subject. The estimation of the F0 is crucial for the classification of the voice signal, because the discrimination of a healthy voice from a pathological one is achieved by evaluating the inclusion of the F0 value within the healthy range. To evaluate the presented methodology, we have carried out a set of tests by using some voice signals selected from an available database in order to compare the classification ability of the proposed methodology with other algorithms existing in the literature. The numerical results obtained show that the proposed methodology provides a good accuracy, sensitivity, and specificity, respectively of over 77%, 72% and 81%, values better than those achieved by the most frequently other used and cited fundamental frequency estimation algorithms. Additionally, a statistical analysis to evaluate whether or not a statistically significant difference exists between the accuracy, sensitivity and specificity has been carried out. The outcome of the ANOVA tests and of the t-tests confirms that there is a significant difference between the proposed methodology and the other algorithms. Finally, the presented methodology could be embedded in a portable and simple m-health application that could be useful for the monitoring of the state of vocal health and the prevention of voice disorders.

Automatic classification possibilities of the voices of children with dysphonia

Dysphonia is a common complaint, almost every fourth child produces a pathological voice. A mobile based filtering system, that can be used by pre-school workers in order to recognize dysphonic voiced children in order to get professional help as soon as possible, would be desired. The goal of this research is to identify acoustic parameters that are able to distinguish healthy voices of children from those with dysphonia voices of children. In addition, the possibility of automatic classification is children. In addition, the possibility of automatic classification is examined. Two sample T-tests were used for statistical significance testing for the mean values of the acoustic parameters between healthy voices and those with dysphonia. A two-class classification was performed between the two groups using leave-one-out cross validation, with support vector machine (SVM) classifier. Formant frequencies, mel-frequency cepstral coefficients (MFCCs), Harmonics-to-Noise Ratio (HNR), Soft Phonation Index (SPI) and frequency band energy ratios, based on intrinsic mode functions measured on different variations of phonemes showed statistical difference between the groups. A high classification accuracy of 93% was achieved by SVM with linear and rbf kernel using only 8 acoustic parameters. Additional data is needed to build a more general model, but this research can be a reference point in the classification of voices using continuous speech between healthy children and children with dysphonia.

Pitch Estimation and Voicing Classification Using Reconstructed Spectrum from MFCC

Pitch Estimation and Voicing Classification Using Reconstructed Spectrum from MFCC

An effective empirical approach to VoIP traffic classification

It is beneficial for telecommunication authorities and Internet service providers (ISPs) to classify and detect voice traffic. It can help them to block unsubscribed users from using their services, which saves them huge revenues. Voice packets can be detected easily, but it becomes complicated when the application or port information in the packet header is hidden due to some secure mechanism such as encryption. This work provides effective voice packet classification and detection based on behavioral and statistical analysis, which is independent of any application, security protocol, or encryption mechanism. First we have made initial assessments through packet feature analysis followed by the implementation of a voice detection algorithm to perform statistical analysis for classifying traffic over IP networks. The proposed voice detection algorithm is executed in three phases: registering of packet flow traces, signature-based analysis, and voice classification. In the first phase, new packets are registered. In the second phase, registered packets are tested if they are already marked as detected. In the third phase, the voice detection algorithm works at distinguishing encrypted and nonencrypted voice flows by fine-tuning the parameters, which are chosen after a detailed statistical analysis of datasets on security protocols such as secure socket layer, secure session initiation protocol, and secure real-time transport protocol. Our results demonstrate a high true positive rate (TPR) and very low false alarm rate (FAR). The proposed methodology achieves a TPR of 93.6% for offline traces, 100% for the self-configured voice setups, and 95% for the online traffic. The FAR is 0.000084% for offline traces and 0.00020% for online traces, which shows that the proposed methodology is highly efficient and can be incorporated in contemporary telecommunication systems.

Morphometric Differences of Vocal Tract Articulators in Different Loudness Conditions in Singing.

IntroductionDynamic MRI analysis of phonation has gathered interest in voice and speech physiology. However, there are limited data addressing the extent to which articulation is dependent on loudness.Material and Methods12 professional singer subjects of different voice classifications were analysed concerning the vocal tract profiles recorded with dynamic real-time MRI with 25fps in different pitch and loudness conditions. The subjects were asked to sing ascending scales on the vowel /a/ in three loudness conditions (comfortable = mf, very soft = pp, very loud = ff, respectively). Furthermore, fundamental frequency and sound pressure level were analysed from the simultaneously recorded optical audio signal after noise cancellation.ResultsThe data show articulatory differences with respect to changes of both pitch and loudness. Here, lip opening and pharynx width were increased. While the vertical larynx position was rising with pitch it was lower for greater loudness. Especially, the lip opening and pharynx width were more strongly correlated with the sound pressure level than with pitch.ConclusionFor the vowel /a/ loudness has an effect on articulation during singing which should be considered when articulatory vocal tract data are interpreted.

A hybrid model for neurological disordered voice classification using time and frequency domain features

Different neurological disorders may lead to speech related problems, due to paralysis in vocal fold or weakness of the related muscles. This may modify the acoustic characteristics of the subject’s voice which may provide important information for detecting certain neurological diseases. The vowel phonation which is acoustically informative and uttered by the patient with not much difficulty is collected and various acoustic features are extracted by time domain and frequency domain techniques. The use of all these features for classification may lead to a large feature space, which may lead to complexity. Hence to avoid this, in the present work experimentation is done by fusing different classifiers which are fed with features extracted from different domains. The time domain and frequency domain features are given to Support Vector Machine (SVM) and Gaussian Mixture Model (GMM) respectively, and the intermediate decision of these classifiers is given to another SVM to identify the voice signal as normal or diseased. It is observed that this hybrid classifier model has shown some improvement with a classification accuracy of 91.43% compared to single GMM classifier with an accuracy of classification of 90% with frequency domain features as input.

Reconstruction-based speech enhancement from robust acoustic features

This paper proposes a method of speech enhancement where a clean speech signal is reconstructed from a sinusoidal model of speech production and a set of acoustic speech features. The acoustic features are estimated from noisy speech and comprise, for each frame, a voicing classification (voiced, unvoiced or non-speech), fundamental frequency (for voiced frames) and spectral envelope. Rather than using different algorithms to estimate each parameter, a single statistical model is developed. This comprises a set of acoustic models and has similarity to the acoustic modelling used in speech recognition. This allows noise and speaker adaptation to be applied to acoustic feature estimation to improve robustness. Objective and subjective tests compare reconstruction-based enhancement with other methods of enhancement and show the proposed method to be highly effective at removing noise.

Voice data mining for laryngeal pathology assessment

The aim of this study was to evaluate the usefulness of different methods of speech signal analysis in the detection of voice pathologies. Firstly, an initial vector was created consisting of 28 parameters extracted from time, frequency and cepstral domain describing the human voice signal based on the analysis of sustained vowels /a/, /i/ and /u/ all at high, low and normal pitch. Afterwards we used a linear feature extraction technique (principal component analysis), which enabled a reduction in the number of parameters and choose the most effective acoustic features describing the speech signal. We have also performed non-linear data transformation which was calculated using kernel principal components. The results of the presented methods for normal and pathological cases will be revealed and discussed in this paper. The initial and extracted feature vectors were classified using the k-means clustering and the random forest classifier. We found that reasonably good classification accuracies could be achieved by selecting appropriate features. We obtained accuracies of up to 100% for classification of healthy versus pathology voice using random forest classification for female and male recordings. These results may assist in the feature development of automated detection systems for diagnosis of patients with symptoms of pathological voice.

The Poultry Voice Classification Model Based on EMD and Support Vector Machine

According to the non-stationary and non-linear characteristics of poultry voice and the situation that it`s hard to obtain enough sound samples, a poultry voice classification method based on Empirical Mode Decomposition (EMD), Teager energy transformation, and Support Vector Machine (SVM) is proposed. Firstly, the poultry voice signals are decomposed into a finite number of intrinsic mode function (IMF).Then, the Teager energy of five IMFs filtered are used to form characteristic vectors. Finally, the eigenvectors are put into a support vector machine classifier . The results of animal voice signals experimental recognition showed that this method had high accuracy and good generalization abilities even in the case of small number of samples. The approach proposed could identify the poultry voice effectively.

Towards developing a voice pathologies detection system

Most voice features used in predicting the voice when a person has voted with instability in the vocal fold vibration cause problems in estimating such period; as a result of this challenge, scientists have focused on the development of powerful features independent of pitch estimation. The major goal of this paper is to study and investigate the Acoustic Voice Analysis methods (AVA) based on adaptive features. This investigation will lead to the development of a system of detection. The essential parts of this topic is related to database (described later), sampling the sounds (and satisfying) from the German database with many diseases, degenerative neurological disease (such as chronic inflammation of the larynx and vocal fold nodules). Under the supervision of the used algorithm to accomplish the above task, the Mel-Frequency Cepstral Coefficients (MFCCs with different Jitter and Shimmer), as by likely flux model mixture (GMM) are used in the AVA. MATLAB was used to simulate such a study for the extraction of features as well as making the training and testing process. The achieved results showed that with some kind of analysis, it is possible to find different sound patterns of diseases, e.g. excessive twang, where additional spectral components exist due to the increase in air flow in nasal cavities. Another focal point is some mathematical transformations both in the temporal domain or frequency. These changes can improve the capacity of some voice features voice; however, there is a need to multivariate analysis of parameters which measure the various problems in the process of phonation; after that, it is necessary to analyse the importance of finding and sorting those features that provide more information. Finally, automatic classification of pathological voices was made using any of the known techniques for this purpose. Our achieved results prove that a good classification rate needs efficient features to characterize each class, in this work, on one hand the accuracy of system increases with the number of parameters (best accuracy with 39 coefficients including Jitter & Shimmer) which means that the difference between normal and abnormal become noticeable with second derivate of MFCC and energy more than the others.

A Pathological Voices Assessment Using Classification

The diagnosing of pathological voice may be a tedious topic and it receives abundant attention. There are several diseases that adversely have an effect on our human speech (voice). The doctor will use only the equipments for detection of pathological voice. However, it is invasive and needs a skilled analysis of diverse human speech signal parameters. Automatic voice analysis for pathological speech has its own blessings, like 1)its quantitative and non-invasive nature. 2) permitting the identification and observance of vocal system diseases . Within the pathological voice classification techniques gathered by the voice of a patient, the goal is to discriminate whether the given voice is normal or pathological. From the speech Mel- Frequency Cepstral Coefficients (MFCC) has been extracted from the voice information and classified into two categories. However, the accuracy of the earlier classification methodology may need additional improvement. In my project work, Support Vector Machine (SVM) classifier is used for pathological voice classification with non-invasive nature to diagnose and analyze the voice of the patient.

A Pathological Voices Assessment Using Classification

The diagnosing of pathological voice may be a tedious topic and it receives abundant attention. There are several diseases that adversely have an effect on our human speech (voice). The doctor will use only the equipments for detection of pathological voice. However, it is invasive and needs an skilled analysis of diverse human speech signal parameters. Automatic voice analysis for pathological speech has its own blessings, like 1) its quantitative and non-invasive nature. 2) permitting the identification and observance of vocal system diseases. Within the pathological voice classification techniques gathered by the voice of a patient, the goal is to discriminate whether the given voice is normal or pathological. From the speech Mel-Frequency Cepstral Coefficients (MFCC) has been extracted from the voice information and classified into two categories. However, the accuracy of the earlier classification methodology may need additional improvement. In my project work, Support Vector Machine (SVM) classifier is used for pathological voice classification with non-invasive nature to diagnose and analyze the voice of the patient.

The Voice of Emotion across Species: How Do Human Listeners Recognize Animals' Affective States?

Voice-induced cross-taxa emotional recognition is the ability to understand the emotional state of another species based on its voice. In the past, induced affective states, experience-dependent higher cognitive processes or cross-taxa universal acoustic coding and processing mechanisms have been discussed to underlie this ability in humans. The present study sets out to distinguish the influence of familiarity and phylogeny on voice-induced cross-taxa emotional perception in humans. For the first time, two perspectives are taken into account: the self- (i.e. emotional valence induced in the listener) versus the others-perspective (i.e. correct recognition of the emotional valence of the recording context). Twenty-eight male participants listened to 192 vocalizations of four different species (human infant, dog, chimpanzee and tree shrew). Stimuli were recorded either in an agonistic (negative emotional valence) or affiliative (positive emotional valence) context. Participants rated the emotional valence of the stimuli adopting self- and others-perspective by using a 5-point version of the Self-Assessment Manikin (SAM). Familiarity was assessed based on subjective rating, objective labelling of the respective stimuli and interaction time with the respective species. Participants reliably recognized the emotional valence of human voices, whereas the results for animal voices were mixed. The correct classification of animal voices depended on the listener's familiarity with the species and the call type/recording context, whereas there was less influence of induced emotional states and phylogeny. Our results provide first evidence that explicit voice-induced cross-taxa emotional recognition in humans is shaped more by experience-dependent cognitive mechanisms than by induced affective states or cross-taxa universal acoustic coding and processing mechanisms.

Effective dysphonia detection using feature dimension reduction and kernel density estimation for patients with Parkinson's disease.

Detection of dysphonia is useful for monitoring the progression of phonatory impairment for patients with Parkinson’s disease (PD), and also helps assess the disease severity. This paper describes the statistical pattern analysis methods to study different vocal measurements of sustained phonations. The feature dimension reduction procedure was implemented by using the sequential forward selection (SFS) and kernel principal component analysis (KPCA) methods. Four selected vocal measures were projected by the KPCA onto the bivariate feature space, in which the class-conditional feature densities can be approximated with the nonparametric kernel density estimation technique. In the vocal pattern classification experiments, Fisher’s linear discriminant analysis (FLDA) was applied to perform the linear classification of voice records for healthy control subjects and PD patients, and the maximum a posteriori (MAP) decision rule and support vector machine (SVM) with radial basis function kernels were employed for the nonlinear classification tasks. Based on the KPCA-mapped feature densities, the MAP classifier successfully distinguished 91.8% voice records, with a sensitivity rate of 0.986, a specificity rate of 0.708, and an area value of 0.94 under the receiver operating characteristic (ROC) curve. The diagnostic performance provided by the MAP classifier was superior to those of the FLDA and SVM classifiers. In addition, the classification results indicated that gender is insensitive to dysphonia detection, and the sustained phonations of PD patients with minimal functional disability are more difficult to be correctly identified.

Classification of Normal and Pathological Voice Using SVM and RBFNN

The identification and classification of pathological voice are still a challenging area of research in speech processing. Acoustic features of speech are used mainly to discriminate normal voices from pathological voices. This paper explores and compares various classification models to find the ability of acoustic parameters in differentiating normal voices from pathological voices. An attempt is made to analyze and to discriminate pathological voice from normal voice in children using different classification methods. The classification of pathological voice from normal voice is implemented using Support Vector Machine (SVM) and Radial Basis Functional Neural Network (RBFNN). The normal and pathological voices of children are used to train and test the classifiers. A dataset is constructed by recording speech utterances of a set of Tamil phrases. The speech signal is then analyzed in order to extract the acoustic parameters such as the Signal Energy, pitch, formant frequencies, Mean Square Residual signal, Reflection coefficients, Jitter and Shimmer. In this study various acoustic features are combined to form a feature set, so as to detect voice disorders in children based on which further treatments can be prescribed by a pathologist. Hence, a successful pathological voice classification will enable an automatic non-invasive device to diagnose and analyze the voice of the patient.

Packet Loss Concealment Algorithm Based on Robust Voice Classification in Noise Environment

실시간 VoIP 네트워크는 지연, 지터 그리고 패킷손실과 같은 네트워크 장애요소로 인해 품질저하가 발생한다. 본 논문은 VoIP 음질 향상을 위해 잡음환경에 강인한 음성분류기반의 패킷손실 은닉 알고리즘을 제안한다. 제안된 방식에서는 음성신호로부터 추출된 다양한 특징들을 분석하고 이를 기반으로 획득된 적응적인 문턱값을 사용하여 수신단에 도착한 패킷을 분류한다. 정확한 신호분류 결과는 패킷손실 은닉에 사용된다. 그리고 선형 예측 기반의 손실패킷 은닉은 연속적으로 패킷을 은닉하거나 손실된 패킷복원 시 발생하는 메탈릭 아티펙트를 제거함으로써 고품질의 음성을 제공한다. The quality of real-time Voice over Internet Protocol (VoIP) network is affected by network impariments such as delays, jitters, and packet loss. This paper proposes a packet loss concealment algorithm based on voice classification for enhancing VoIP speech quality. In the proposed method, arriving packets are classified by an adaptive thresholding approach based on the analysis of multiple features of short signal segments. The excellent classification results are used in the packet loss concealment. Additionally, linear prediction-based packet loss concealment delivers high voice quality by alleviating the metallic artifacts due to concealing consecutive packet loss or recovering lost packet.

Classification System of Pathological Voices Using Correntropy

This paper proposes the use of a similarity measure based on information theory called correntropy for the automatic classification of pathological voices. By using correntropy, it is possible to obtain descriptors that aggregate distinct spectral characteristics for healthy and pathological voices. Experiments using computational simulation demonstrate that such descriptors are very efficient in the characterization of vocal dysfunctions, leading to a success rate of 97% in the classification. With this new architecture, the classification process of vocal pathologies becomes much more simple and efficient.