Source-filter Comparison of Measurements of Fundamental Frequency Perturbation and Amplitude Perturbation for Synthesized Voice Signals

SUMMARY: An investigation of the effect of glottal source aperiodicities (jitter, shimmer, and aspiration noise) on the estimation of fundamental frequency (f0) perturbation and amplitude perturbation, of synthesized, glottal source and voiced speech waveforms, is considered. Firstly, 4, cycle-event f0 estimators are examined: (1) waveform matching of the low-pass filtered waveform, (2) positive peaks (PPs) from the speech waveform, (3) PPs from the low-pass filtered waveform, and (4) positive zero crossings from the low-pass filtered waveform. The analysis shows that f0 perturbation measures taken from the low-pass filtered waveform are affected by both amplitude perturbation and random glottal noise, whereas, f0 perturbation measures taken from the PPs of the original waveform are affected by noise but not by amplitude perturbation. It is shown for the low-pass filter methods that the effects of amplitude perturbation and noise lead to increased errors in the measurement of f0 perturbation for the synthesized speech waveforms when compared with the synthesized glottal waveforms. Shimmer of the synthesized speech waveform is approximately equal to shimmer of the synthesized glottal source. However, noise and jitter affect measures of amplitude perturbation. The estimation of f0 perturbation from the synthesized speech waveform is shown to be nonlinearly related to f0 perturbation estimation from the synthesized glottal waveform as a consequence of the filtering action of the vocal tract. Low-pass filtering the voiced speech waveform is shown to provide a partial solution to this problem.     

扩散噪声下协方差矩阵重构的语音分离与降噪

针对传统多通道语音分离算法在扩散噪声下性能下降的问题,提出了一种用于语音分离及降噪的空间协方差模型及参数估计方法。该方法将扩散噪声视为独立声源,利用由导向矢量重构的空间协方差矩阵建模目标声源的空间特性,并通过空间协方差分析方法估计用于语音分离的多通道维纳滤波器。同时,还提出了一种联合该方法的后置滤波器参数框架,为输出信号降噪和失真的折中提供了更多选择。在扩散噪声下的单目标和多目标实验中,所提方法的语音提取和分离性能都优于对比算法,联合参数的后置滤波器可提供更为符合人们要求的降噪语音,验证了所提模型与参数估计方法的有效性。      

Effects of lung volume on the glottal voice source

According to experience in voice therapy and singing pedagogy, breathing habits can be used to modify phonation, although this relationship has never been experimentally demonstrated. In the present investigation we examine if lung volume affects phonation. Twenty-four untrained subjects phonated at different pitches and degrees of vocal loudness at different lung volumes. Mean subglottal pressure was measured and voice source characteristics were analyzed by inverse filtering. The main results were that with decreasing lung volume, the closed quotient increased, while subglottal pressure, peak-to-peak flow amplitude, and glottal leakage tended to decrease. In addition, some estimates of the amount of the glottal adduction force component were examined. Possible explanations of the findings are discussed.     

Accuracy in determining voice source parameters

The paper addresses the accuracy of an approximate solution to the inverse problem of retrieving the shape of a voice source from a speech signal for a known signal-to-noise ratio (SNR). It is shown that if the source is found as a function of time with the A.N. Tikhonov regularization method, the accuracy of the found approximation is worse than the accuracy of speech signal recording by an order of magnitude. In contrast, adequate parameterization of the source ensures approximate solution accuracy comparable with the accuracy of the problem data. A corresponding algorithm is considered. On the basis of linear (in terms of data errors) estimates of approximate parametric solution accuracy, parametric models with the best accuracy can be chosen. This comparison has been carried out for the known voice source models, i.e., model [17] and the LF model [18]. The advantages of the latter are shown. Thus, for SNR = 40 dB, the relative accuracy of an approximate solution found with this algorithm is about 1% for the LF model and about 2% for model [17] as compared to an accuracy of 7–8% in the regularization method. The role of accuracy estimates found in speaker identification problems is discussed.     

SIM--simultaneous inverse filtering and matching of a glottal flow model for acoustic speech signals

A new method "simultaneous inverse filtering and model matching" (SIM) is proposed that allows one to calculate voice source measures without any user interaction. It is based on the discrete all-pole modeling (DAP) technique for inverse filtering (IF), which is modified to include a model of the glottal flow as integral part [LF model, Fant et al., STL-QPSR (Stockholm) 4/1985, 1-13 (1986)]. As the correct LF parameters are initially unknown, they are estimated in an iterative procedure using multi-dimensional optimization techniques that are initialized according to the results of an exhaustive search. The error criteria applied reflect how well the IF is performed after the spectral contribution of the glottal flow has been removed. The resulting optimal LF parameter constellation serves as the basis to calculate 11 voice source measures. The performance was evaluated using synthesized signals and recordings of natural utterances. For the synthesized signals, the accuracy to reproduce the original parameters was high (correlations exceeding 0.88) for measures where the starting point of the glottal cycle did not enter explicitly. Errors were smaller compared to conventional estimation methods where the measures were estimated from the IF signal. The analysis of natural utterances indicates that problems still exist with regard to robustness, but that under advantageous conditions the open quotient, the speed quotient, the closing quotient, the parabolic spectral parameter, and the negative peak amplitude of the glottal flow derivative can indeed be determined automatically by the SIM method.     

The contribution of the excitatory source to the perception of neutral vowels in stuttered speech

The vowel in part-word repetitions in stuttered speech often sounds neutralized. In the present article, measurements of the excitatory source made during such episodes of dysfluency are reported. These measurements show that, compared with fluent utterances, the glottal volume velocities are lower in amplitude and shorter in duration and that the energy occurs more towards the low-frequency end of the spectrum. In a first perceptual experiment, the effects of varying the amplitude and duration of the glottal source were assessed. The glottal volume velocity recordings of the /ae/ vowels used in the analyses were employed as driving sources for an articulatory synthesizer so that judgments about the vowel quality could be made. With dysfluent glottal sources (either as spoken or by editing a fluent source so that it was low in amplitude and brief), the vowels sounded more neutralized than with fluent glottal sources (as spoken or by editing a dysfluent source to increase its amplitude and lengthen it). In a second perceptual experiment, synthetic glottal volume velocities were used to verify these findings and to assess the influence of the low-frequency emphasis in the dysfluent speech. This experiment showed that spectral bias and duration both cause stuttered vowels to sound neutralized.     

Segmentation of the period of the fundamental tone of a voice source

The extrema of the logarithmic derivative of the mean energy of a voice signal in the frequency range of 1000–3000 Hz are used to determine the instants of opening and closure of the glottis. The inaccuracy of analysis is estimated with the Arctic CMU database, which contains synchronous recordings of speech signals and electro-glottograms. The estimates of the instants of opening and closure of the glottis, found by the developed algorithm, are compared with the instants of the maximum and minimum of the derivative from electro-glottogram signals, which are taken as the “true” instants. The mean square deviation of the glottal opening instant from the extrema of the derivative from the electro-glottogram signals for different speakers is in the range of 1.03–1.64 ms. The error rate of a false estimate of the glottal opening instant is from 0.01 to 0.14%, and the error rate of omission is from 0.42 to 2.38%. An error-detection algorithm is developed. The mean square deviation with an relative—to the period of the fundamental tone—error in detecting the glottal opening instant is in the range of 13–18% for the most probable error from 0 to +5%.     

Age, sex, and vowel dependencies of acoustic measures related to the voice source

The effects of age, sex, and vocal tract configuration on the glottal excitation signal in speech are only partially understood, yet understanding these effects is important for both recognition and synthesis of speech as well as for medical purposes. In this paper, three acoustic measures related to the voice source are analyzed for five vowels from 3145 CVC utterances spoken by 335 talkers (8-39 years old) from the CID database [Miller et al., Proceedings of ICASSP, 1996, Vol. 2, pp. 849-852]. The measures are: the fundamental frequency (F0), the difference between the "corrected" (denoted by an asterisk) first two spectral harmonic magnitudes, H1* - H2* (related to the open quotient), and the difference between the "corrected" magnitudes of the first spectral harmonic and that of the third formant peak, H1* - A3* (related to source spectral tilt). The correction refers to compensating for the influence of formant frequencies on spectral magnitude estimation. Experimental results show that the three acoustic measures are dependent to varying degrees on age and vowel. Age dependencies are more prominent for male talkers, while vowel dependencies are more prominent for female talkers suggesting a greater vocal tract-source interaction. All talkers show a dependency of F0 on sex and on F3, and of H1* - A3* on vowel type. For low-pitched talkers (F0 < or = 175 Hz), H1* - H2* is positively correlated with F0 while for high-pitched talkers, H1* - H2* is dependent on F1 or vowel height. For high-pitched talkers there were no significant sex dependencies of H1* - H2* and H1* - A3*. The statistical significance of these results is shown.     

利用声带动力学模型参数反演方法进行病变嗓音分类

提出一种声带动力学模型参数反演方法,从发声机理角度对声带病变嗓音进行有效区分。依据声带生理组织和伯努利定律构建声带动力学模型,确定模型优化参数向量,耦合声门气流获取模型声门波;利用迭代自适应逆滤波算法获得实际嗓音声门波作为目标声门波;采用遗传优化算法提出通过匹配目标和模型声门波特征参数实现模型参数反演。实验结果表明,表征声门波的各时频域参数匹配相对误差不超过2%;依据反演所获模型参数提出去除声门下压影响的平均归一化缩放系数,克服声带非对称性特征在区分病变嗓音方面的不足,实现病理嗓音的全面有效区分。      

Vocal quality factors: analysis, synthesis, and perception.

The purpose of this study was to examine several factors of vocal quality that might be affected by changes in vocal fold vibratory patterns. Four voice types were examined: modal, vocal fry, falsetto, and breathy. Three categories of analysis techniques were developed to extract source-related features from speech and electroglottographic (EGG) signals. Four factors were found to be important for characterizing the glottal excitations for the four voice types: the glottal pulse width, the glottal pulse skewness, the abruptness of glottal closure, and the turbulent noise component. The significance of these factors for voice synthesis was studied and a new voice source model that accounted for certain physiological aspects of vocal fold motion was developed and tested using speech synthesis. Perceptual listening tests were conducted to evaluate the auditory effects of the source model parameters upon synthesized speech. The effects of the spectral slope of the source excitation, the shape of the glottal excitation pulse, and the characteristics of the turbulent noise source were considered. Applications for these research results include synthesis of natural sounding speech, synthesis and modeling of vocal disorders, and the development of speaker independent (or adaptive) speech recognition systems.     

Automatic Classification of Healthy Subjects and Patients With Essential Vocal Tremor Using Probabilistic Source-Filter Model Based Noise Robust Pitch Estimation

Essential voice tremor (EVT) is a voice disorder resulting from dyscoordination within the laryngeal musculature. A low-frequency fluctuations of fundamental voice frequency or the strength of excitation amplitude is the main consequence of the disorder. The automatic classification of healthy control and EVT is useful tool for the clinicians. A typical automatic EVT classification involves three steps. The first step is to compute the pitch contour from the speech. The second step is to compute the features from the pitch contour, and the final step is to use a classifier to classify the features into healthy or EVT. It is shown that a high-resolution pitch contour estimated from the glottal closure instants (GCIs) is useful for EVT classification. The HPRC estimation can be very poor in the presence of noise. Hence, a probabilistic source filter model based noise robust GCI detection is used for HPRC estimation. The Empirical mode decomposition based feature extraction is used followed by a support vector machine classifier. The EVT classification performance is evaluated using recordings from 45 subjects. The proposed method is found to perform better than the baseline techniques in eight different additive noise conditions with six SNR levels.     

A flow waveform-matched low-dimensional glottal model based on physical knowledge

The purpose of this study is to explore the possibility for physically based mathematical models of the voice source to accurately reproduce inverse filtered glottal volume-velocity waveforms. A low-dimensional, self-oscillating model of the glottal source with waveform-matching properties is proposed. The model relies on a lumped mechano-aerodynamic scheme loosely inspired by the one- and multimass lumped models. The vocal folds are represented by a single mechanical resonator and a propagation line which takes into account the vertical phase differences. The vocal-fold displacement is coupled to the glottal flow by means of an aerodynamic driving block which includes a general parametric nonlinear component. The principal characteristics of the flow-induced oscillations are retained, and the overall model is able to match inverse-filtered glottal flow signals. The method offers in principle the possibility of performing transformations of the glottal flow by acting on the physiologically based parameters of the model. This is a desirable property, e.g., for speech synthesis applications. The model was tested on a data set which included inverse-filtered glottal flow waveforms of different characteristics. The results demonstrate the possibility of reproducing natural speech waveforms with high accuracy, and of controlling important characteristics of the synthesis such as pitch.     

Acoustic analysis of trill sounds

In this paper, the acoustic-phonetic characteristics of steady apical trills--trill sounds produced by the periodic vibration of the apex of the tongue--are studied. Signal processing methods, namely, zero-frequency filtering and zero-time liftering of speech signals, are used to analyze the excitation source and the resonance characteristics of the vocal tract system, respectively. Although it is natural to expect the effect of trilling on the resonances of the vocal tract system, it is interesting to note that trilling influences the glottal source of excitation as well. The excitation characteristics derived using zero-frequency filtering of speech signals are glottal epochs, strength of impulses at the glottal epochs, and instantaneous fundamental frequency of the glottal vibration. Analysis based on zero-time liftering of speech signals is used to study the dynamic resonance characteristics of vocal tract system during the production of trill sounds. Qualitative analysis of trill sounds in different vowel contexts, and the acoustic cues that may help spotting trills in continuous speech are discussed.     

Investigation of a glottal related harmonics-to-noise ratio and spectral tilt as indicators of glottal noise in synthesized and human voice signals

The harmonics-to-noise ratio (HNR) of the voiced speech signal has implicitly been used to infer information regarding the turbulent noise level at the glottis. However, two problems exist for inferring glottal noise attributes from the HNR of the speech wave form: (i) the measure is fundamental frequency (f0) dependent for equal levels of glottal noise, and (ii) any deviation from signal periodicity affects the ratio, not just turbulent noise. An alternative harmonics-to-noise ratio formulation [glottal related HNR (GHNR')] is proposed to overcome the former problem. In GHNR' a mean over the spectral range of interest of the HNRs at specific harmonic/between-harmonic frequencies (expressed in linear scale) is calculated. For the latter issue [(ii)] two spectral tilt measures are shown, using synthesis data, to be sensitive to glottal noise while at the same time being comparatively insensitive to other glottal aperiodicities. The theoretical development predicts that the spectral tilt measures reduce as noise levels increase. A conventional HNR estimator, GHNR' and two spectral tilt measures are applied to a data set of 13 pathological and 12 normal voice samples. One of the tilt measures and GHNR' are shown to provide statistically significant differentiating power over a conventional HNR estimator.     

Voice source characteristics in six premier country singers

Voice source characteristics as derived from inverse filtering were analyzed in 6 country singers' speech and singing. Results showed that the closed quotient varied systematically with vocal loudness, and that glottal compliance (the ratio between transglottal AC volume displacement and subglottal pressure) decreased with increases in fundamental frequency but remained unaffected by vocal loudness. No striking differences were found in source characteristics between speech and singing within subjects. The degree of phonatory press, as judged by a panel of 19 expert listeners, appeared related to the range in which the singer was singing and to the sound pressure level gain from a doubling of subglottal pressure.     

Pressure measurements during speech production using semiconductor miniature pressure transducers: impact on models for speech production

It appears that temperature instabilities are a major obstacle hindering the use of semiconductor strain gauge pressure transducers in speech research, especially when absolute pressure data are mandatory. In this paper a simple and reliable method for an in vivo calibration of this kind of transducer is described. The most important error source, the drift of the zero pressure level due to temperature changes, is discussed, and an estimation of the measurement accuracy which can be obtained is given. Moreover, some registrations of subglottal, supraglottal, and transglottal pressure are presented. It is shown that the pressure recordings allow us to obtain estimates of the volume flow in the trachea and pharynx. Analysis of those waveforms appears to lead to new insights into the physical processes underlying voice production. Specifically, an independent glottal contribution to the skewing of the glottal flow pulses is identified.     

Spectral characterization of jitter, shimmer, and additive noise in synthetically generated voice signals

Alteration of the harmonic structure in voice source spectra, taken over at least two periods of the waveform, may occur due to the presence of fundamental frequency (f0) perturbation, amplitude perturbation, additive noise, or changes within the glottal source signal itself. In order to make accurate inferences regarding glottal-flow dynamics or perceptual evaluations based on spectral measurements taken from the acoustic speech waveform, investigation of the spectral features of each aperiodic component is required. Based on a heuristic development involving a consideration of the partial sum of the Fourier series taken for two periods of a jittered, shimmered, and (additive, random) noise-contaminated signal, the corresponding spectral characteristics are hypothesized. Subsequent to this, the Fourier series coefficients are calculated for the two periods in order to test the hypotheses. Definite spectral differences are found for each aperiodic component; based on these findings differential quantitative spectral measurements are suggested. Further supportive evidence is obtained through use of Fourier transform and periodogram-averaged calculations. The analysis is carried out on synthetically generated glottal-pulse waveforms and on radiated speech waveforms. A discussion of the results is given in terms of voice aperiodicity in general and in terms of their implication for future studies involving human voice signals.     

Comparison of Two Inverse Filtering Methods in Parameterization of the Glottal Closing Phase Characteristics in Different Phonation Types

SUMMARY: Inverse filtering (IF) is a common method used to estimate the source of voiced speech, the glottal flow. This investigation aims to compare two IF methods: one manual and the other semiautomatic. Glottal flows were estimated from speech pressure waveforms of six female and seven male subjects producing sustained vole /a/ in breathy, normal, and pressed phonation. The closing phase characteristics of the glottal pulse were parameterized using two time-based parameters: the closing quotient (C1Q) and the normalized amplitude quotient (NAQ). The information given by these two parameters indicates a strong correlation between the two IF methods. The results are encouraging in showing that the parameterization of the voice source in different speech sounds can be performed independently of the technique used for inverse filtering.