This paper describes the implementation of a waveform-based global dynamic model with the goal of capturing vocal folds variability. The residue extracted from speech by inverse filtering is pre-processed to remove phoneme dependence and is used as the input time series to the dynamic model. After training, the dynamic model is seeded with a point from the trajectory of the time series, and iterated to produce the synthetic excitation waveform. The output of the dynamic model is compared with the input time series. These comparisons confirmed that the dynamic model had captured the variability in the residue. The output of the dynamic models is used to synthesize speech using a pitch-synchronous speech synthesizer, and the output is observed to be close to natural speech.

The paper shows that an adaptive weighted recursive least squares algorithm with a variable forgetting factor (WRLS-VFF) will adjust the size of the data segment to be analyzed according to its time-varying characteristics, as during the transitions between vowels and consonants. The algorithm can accurately estimate the vocal tract formants, anti-formants, and their bandwidths, be used for glottal inverse filtering, perform voiced (V)/unvoiced (U)/silent (S) classification of speech segments, estimate the input excitation (either white noise or periodic pulse trains), and estimate the instant of glottal closure.>

An optimization approach for estimating vocal tract configurations from the speech signal is presented. The scheme concatenates a gradient search and linear successive approximation, providing fast convergence and small errors. The objective function is generated by the least-absolute-value (l/sub 1/-norm) error between the model-derived and the speech-derived first three formants. The gradient search is accelerated by using an algorithm inspired by the Fletcher-Reeves method. Linear successive approximation assures convergence in the region near the optimum articulatory vector. Constraints are imposed on the articulatory parameters to avoid physiologically impossible configurations.>

The authors consider the synthesis of speech with various vocal characteristics, such as modal, creaky, breathy, rough, and hoarse. The relationships between the acoustical parameters of the glottal source pulses and these vocal characteristics were reviewed. Based on the acoustical parameters needed to model these vocal characteristics, a novel unified glottal source model was developed. This source model was tested with a formant synthesizer for its ability to synthesize speech with various vocal characteristics. Informal listening tests were conducted to evaluate the performance of this source model. The preliminary results are encouraging, and more rigorous testing of this model and further development of the glottal source model for each vocal characteristic are underway. Applications of this work include high-quality speech synthesis and the modeling of vocal disorders.>

The authors have modified Klatt's formant synthesizer and have implemented it in software. The user can configure the filter bank(s) by simple parameter specifications. This flexible formant synthesizer has been tested by synthesizing segments of sounds with a variable number of formant tracks and also with unsmooth and discontinuous formant tracks. This synthesizer can be used as tool in experiments with speech analysis-by-synthesis and rule-based speech synthesis.>

A weighted recursive least-squares algorithm with a variable forgetting factor (WRLS-VFF) is introduced for speech signal analysis. The variable forgetting factor, which indicates the state change of the estimator, can be used to estimate the input excitation when the input is either white noise or periodic pulse trains. Two analysis techniques are examined: glottal closed-phase adaptive formant tracking and glottal closed-phase inverse filtering. The glottal closed-phase interval can be located approximately from the VFF estimation error. The data analyzed include synthesized speech segments and isolated words and sentences from real speech. Results show that the WRLS-VFF algorithm offers a more accurate estimation of formants and faster formant tracking than either linear predictive coding or several other adaptive algorithms. In addition, the WRLS-VFF technique is used to obtain, automatically, estimates of the glottal volume-velocity waveform by inverse filtering.>

Two algorithms are given for automatically recognizing the gender of a speaker using acoustic parameters extracted from the speaker's speech. The speech data used for developing the algorithms were taken from a large data set. Only acoustic parameters for vowels and fricatives were used to develop and test the algorithms because the authors wanted the gender classification to be achieved rapidly using only a brief data record.>

A new algorithm is proposed to solve the problem of separating two voices which co-exist on a single channel. This algorithm produces an initial estimate of the speech spectrum for each talker, and then employs multisignal minimum-cross-entropy spectral analysis to improve the initial spectral estimates by accounting for the characteristics of the known composite speech signal. Results of listening tests and spectral distortion comparisons show that the proposed algorithm can recover the desired speech with better intelligibility than other speech separation techniques.

Attempts to measure the synthetic quality of speech usually consider the two factors intelligibility and naturalness, each involving subjective and objective characteristics. To generate high quality synthetic speech, spectral distortion should be avoided, spectral continuity and formant tracking should be done well. Glottal-related factors, including proper modeling of the 1) glottal excitation waveforms and 2) effects of source-tract interaction for synthesizers are discussed. Accurate detection of voiced/unvoiced/ silent segments in the speech waveform and the fundamental frequency of voicing are also major concerns. We present both formal and informal listener evaluations of three synthesizers: LPC, formant and articulatory. Finally, we suggest a two-channel, speech and electroglottograph (EGG), approach to speech analysis to aid the automatic processing of speech.

The electroglottographic (EGG) waveform has been conjectured to be inversely proportional to the lateral area of contact of the vocal folds. This conjecture is difficult to substantiate by direct measurements. We report on a simple model of the vocal folds which we use to estimate the vocal fold lateral contact area from which we calculate the EGG waveform as the reciprocal of the area function. The application of this research to aids for the handicapped is discussed.

The objective of our study is to define and measure source-tract interaction using the speech signal. Numerous researchers have conjectured that the subglottal, glottal, and supraglottal portions of our own speech production mechanism may interact, affecting the quality of our voice. Several methods for incorporating the effects of source-tract interaction into a synthetic speech model have been suggested. Speech synthesized with source-tract interaction sounds more natural than speech generated without such interaction. Can source-tract interaction be parameterized using the speech signal and incorporated into vocoders and speech synthesizers? Can this measurement be accomplished on a pitch period by pitch period basis?

A flexible analysis-synthesis system with signal dependent features is described and used to realize some desired voice characteristics in synthesized speech. The intelligibility of synthetic speech appears to depend on the ability to reproduce dynamic sounds such as stops, whereas the quality of voice is mainly determined by the true reproduction of voiced segments. We describe our work in converting the speech of one speaker to sound like that of another. A number of factors are important for maintaining the quality of the voice during this conversion process. These factors are derived from both the speech and electroglottograph signals.

Accurate characterization of the vocal tract filter relies on speech data obtained from the closed glottal interval. Use of the electroglottograph (EGG) as a noninvasive and reliable means of monitoring glottal vibratory characteristics has increased the popularity of the two channel approach to speech analysis. The cycle by cycle auxiliary glottal information provided by the EGG facilitates implementation of pitch synchronous closed phase procedures. This paper shows that the closed phase method can yield variable results if the EGG information, or an alternative closed phase indicator, is not used correctly for analysis frame positioning. Synchronized glottal area waveforms obtained from ultra high speed laryngeal films verify true closure for our analyses. Results presented for other known regions also confirm many previous observations.

In this paper we describe a flexible analysis-synthesis system which can be used for a number of studies in speech research. The main objective is to have a synthesis system whose characteristics can be controlled through a set of parameters to realize any desired voice characteristics. The basic synthesis scheme consists of two steps: Generation of an excitation signal from pitch and gain contours and excitation of the linear system model described by linear prediction coefficients. We show that a number of basic studies such as time expansion/compression, pitch modifications and spectral expansion/compression can be made to study the effect of these parameters on the quality of synthetic speech. A systematic study is made to determine factors responsible for unnaturalness in synthetic speech. It is found that the shape of the glottal pulse determines the quality to a large extent. We have also made some studies to determine factors responsible for loss of intelligibility in some segments of speech. A signal dependent analysis-synthesis scheme is proposed to improve the intelligibility of dynamic sounds such as stops. A simple implementation of the signal dependent analysis is proposed.

Two channel speech analysis, using the electroglottograph (EGG) and the utterance, is used to parse the utterance into voiced, unvoiced, and silence regions and determine the pitch contour all in real time. Results for two algorithms which use the speech signal alone are compared to our two channel method. The effect a source-tract interaction model, based on our data, has on the production of natural sounding speech is reported. Three glottal source models are compared, namely, Fant's, Guerin's, and the impulse models. The waveforms generated by these models excite a serial/parallel Klatt formant synthesizer to produce synthesized sentences. The major parameters for producing excellent quality synthesized speech are 1) the "shape" of the glottal excitation waveform, 2) the first formant bandwidth, and 3) the locations of the higher (fourth and fifth) formant frequencies.

This paper describes a system for the automatic measurement of numerous vibratory parameters of the vocal cord from high speed films. Our new system, which is an extension of our earlier work, has a much higher gray level range and spatial resolution (256 gray levels, 640×680 pixels) as compared to the previous system (8 gray levels, 90×35 pixels). We present the parameters measured for a typical normal larynx phonating a vowel and compare these parameters with the volume velocity derived by inverse filtering the same utterance by the same subject. This comparison has not been previously reported.

For sometime we have been developing a system to measure parameters of the vibratory patterns of the vocal cords from ultra high speed (∼ 5000 frames/sec) laryngeal films. Previously, we have reported only glottal area measurements, comparing results from our automated system with those obtained by a TV system and by hand measurements using a planimeter and computer based sonic pen. In this paper we report on numerous other vocal cord vibratory parameters measured automatically from films of normal larynges. We are also making similar measurements from films of individuals with laryngeal diseases and other laryngeal malfunctions, but these latter results are not reported here. The objective of this research is to contribute data relevant to a theory of vocal cord motion and to develop aids for the hearing impaired.

Using ideas from one-dimensional maximum entropy spectral estimation a two-dimensional spectral estimator is derived by extrapolating the two-dimensional sampled autocorrelation (or covariance) function. The method used maximizes the entropy of a set of random variables. The extrapolation (or prediction) process under this maximum entropy condition is shown to correspond to the most random extension or equivalently to the maximization of the mean-square prediction error when the optimum predictor is used. The two-dimensional extrapolation must he terminated by the investigator. The Fourier transform of the extrapolated autocorrelation function is the two-dimensional spectral estimator. Using this method one can apply windowing prior to calculating the spectral estimate. A specific algorithm for estimating the two-dimensional spectrum is presented, and its computational complexity is estimated. The algorithm has been programmed and computer examples are presented.

Using the ideas from one-dimensional (1-D) maximum entropy spectral estimation, we derive a 2-D spectral estimator by actually extrapolating the 2-D sampled autocorrelation function. The method used here is to maximize the entropy of a set of random variables. The extrapolation (prediction) process under the maximum entropy condition is shown to correspond to the most random extension or to the maximization of the mean square prediction error conditioned on using the optimum predictor. The 2-D extrapolation must be terminated by the investigator. The Fourier transform of the extrapolated autocorrelation function is our 2-D spectral estimator. A specific algorithm for estimating the 2-D spectrum is presented. The algorithm has been programmed and computer examples are presented.

A progress report is presented on the development of procedures to process laryngeal high speed films to extract the glottal waveform and other glottal parameters. The glottal waveforms derived by this direct method are being used to determine if the pathologic larynx is manifested in its abnormal vibratory pattern. The glottal waveforms measured by sonic-sensing pen tracing, cursor outlining, a photocell technique, and TV camera scanning are presented and compared with the conventional polar planimeter method. To date the TV camera method is the most rapid procedure and can process approximately 400 frames per hour in a semi-automatic mode. This method determines the glottal area, length, and width by a combination of analog/digital circuitry and minicomputer processing.

The relationship between scalp potentials evoked by visual stimulation and neuronal functioning is examined. A neuronal population model for the electrogenesis of the evoked surface potentials is developed which relies on the hypothesis that the surface potentials are a combination of excitatory post-synaptic potentials (EPSPs) and inhibitory post-synaptic potentials (IPSPs) which occur both at different depths and at different latencies. The model is used to determine the location (depth) of the sources and the approximate size of the neuronal population responsible for the potential. The spatio-temporal potential fields evoked by visual stimulation and monitored by an array of scalp electrodes over the occipital cortex are discussed along with our method for displaying these fields via contour maps. The experimental data are computer simulated via discrete spatially disparate sources. The waveform of each of these sources is determined and the various parameters of the waveform are related to the major characteristics of the spatio-temporal visual evoked potential fields as well as to specific features of the underlying neuronal population. These features include, e.g., the sequence of neuronal excitation, the magnitude of excitatory current, the depth of the neuronal population, and the period of activity of the EPSP and IPSP.

A technique for decomposing a composite signal of unknown multiple wavelets overlapping in time is described. The computation algorithm incorporates the power cepstrum and complex cepstrum techniques. It has been found that the power cepstrum is most efficient in recognizing wavelet arrival times and amplitudes while the complex cepstrum is invaluable in estimating the form of the basic wavelet and its echoes, even if the latter are distorted. Digital data-processing problems such as the detection of multiple echoes, various methods of linear filtering the complex cepstrum, the picket-fence phenomenon, minimum-maximum phase situations, and amplitude- versus phase-smoothing for the additive-noise case are examined empirically and where possible theoretically, and are discussed. A similar investigation is performed for some of the preceding problems when the echo or echoes are distorted versions of the wavelet, thereby giving some insight into the complex problem of separating a composite signal composed of several additive stochastic processes. The threshold results are still empirical and the results should be extended to multi-dimensional data. Applications are the decomposition or resolution of signals (e.g., echoes) in radar and sonar, seismology, speech, brain waves, and neuroelectric spike data. Examples of results are presented for decomposition in the absence and presence of noise for specified signals. Results are tendered for the decomposition of pulse-type data appropriate to many systems and for the decomposition of brain waves evoked by visual stimulation.

An algorithm is discussed which decomposes a noisy composite signal of identical but unknown multiple wavelets overlapping in time. The decomposition determines the number of wavelets present, their epochs, amplitudes, and an estimate of the basic wavelet shape. The algorithm is an adaptive decomposition filter which is a combination tion of three separate filters. One is an adaptive cross-correlation filter which resolves the composite signal from noise by an iteration procedure; this is followed by a wavelet extraction filter which ferrets out the basic wavelet form, and last there appears an inverse filter which achieves decomposition of the composite signal in the time domain. The decomposition algorithm can be applied to echoes and overlapping wavelets which might arise in radar, sonar, seismology, or electro-physiology. The proposed theory has been thoroughly simulated and selected experimental results are presented to demonstrate the technique. These include decomposition of brain waves evoked by visual stimulation.

The importance of designing a radar (or communications) receiver to operate effectively in the presence of spatial noise creates the necessity for determining the space-time cross-correlation functions between antenna array elements. A representation of a general, polarized, nonisotropic noise process is used to determine the cross-spectral densities and cross-correlation functions between arbitrarily oriented dipole antenna array elements in the presence of a noise field only. Several examples have been worked out for isotropic noise to show the variation of the cross-correlated noise function vs. the spatial separation of the dipoles. One important result is that the noise cross correlation is shown to be a function of the receiver element spatial orientation when the elements are dipoles. For certain spatial orientations between dipoles it is possible to reduce the noise cross correlation between these antenna array elements to zero.