Vocal Tract Filter Research Articles

Voice source model for continuous control of pitch period.

The voiced speech waveform may be synthesized by exciting an LPC vocal tract filter with a pulse waveform patterned after naturally occurring glottal airflow pulses. Such a pulse waveform may be generated by computing samples of a piecewise polynomial curve at equally spaced time intervals. In this type of synthesis, the pitch period is commonly restricted to an integer multiple of the sample interval. A method is presented for removing this restriction, permitting both pulse duration and pitch period to be varied over continuous time. Aliasing distortion is prevented by computing the sample values of pulses that have been low-pass filtered in continuous time prior to sampling. Applications of this technique include modeling glottal pulses by least-squares fit to inverse filter waveforms, the synthesis of calibration waveforms for evaluating measures of speech waveform jitter, the perceptual evaluation of low levels of waveform jitter, and the synthesis of the singing voice.

Contributions of voice-source and vocal-tract characteristics to speaker identity

A study was done on whether a voice-source model, in particular the Liljencrants–Fant model, codes information that is used by listeners to identify speakers by their voices. Automatic analysis/resynthesis techniques were used to generate so-called hybrid vowels for which the voice-source characteristics of one speaker are combined with the vocal-tract characteristics of another speaker. Listeners who were trained to recognize the speakers by their natural utterances, had to indicate which speaker they thought had produced the hybrid stimulus. It was found that there is no general rule that vocal-tract information contributes more to perceived speaker identity than voice-source information. Sometimes vocal-tract information is more important, sometimes voice-source information. The results fit the subjective cues that listeners reported they had been using to perform the speaker identification task quite well. Besides the expected importance of the vocal-tract filter, the spectral balance between high- and low-frequency components of the voice-source spectrum and the flutter of F0 proved to be important perceptual cues for speaker identity.

Glottal source estimation using a sum-of-exponentials model

An algorithm for estimating the glottal source waveform in voiced speech is described. The glottal source waveform is described using the LF model proposed by Fant et al. (1985). The vocal tract filter is modeled as a pole-zero system. The analysis of vowel sounds from several talkers shows that the analysis procedure leads to an accurate estimate of the glottal source.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Glottal source estimation using a sum of exponentials model

This paper describes an algorithm for simultaneously estimating the parameters of a model for the glottal source and the vocal tract filter. The glottal source signal is described by t.he four‐parameter LF model [Fant et al., Speech Transmis. Lab. Q. Prog. Stat. Rep. (1984)]. The vocal tract is modeled as an all‐pole filter; however, to allow for the effects of source‐tract interaction, a separate model is used in the closed and open glottal phases. Under these assumptions, it is shown that the output speech signal can be modeled as a sum of complex exponential signals. The parameters of the speech signal model are estimated using the algorithms described in Kumaresan and Tufts [IEEE Trans. Acoust. Speech Signal Process. ASSP‐30, 833–840 (1982)]. The electroglottograph signal is used to obtain a preliminary estimate of the location of the closed and open phases in each pitch period. The analysis of steady vowel sounds from several speakers indicates that this method provides a very accurate estimate of th...

Estimating the acoustic characteristics of speech from visual speech signals

Speech articulation produces acoustic and visual signals. When the acoustic signal is degraded by noise, the visual signal can provide compensatory information [W. H. Sumby and I. Pollack, J. Acoust. Soc. Am. 26, 212–215 (1954)]. The aim of the work to be presented is to explore the extent to which the transfer function of the vocal tract filter can be estimated from the visual speech signal. The approach was to use a neural network to obtain the short-term power spectral envelope of the acoustic signal given the corresponding visual image as input. The visual images were taken from laser disc recordings of speaker's faces [Bernstein et al., J. Acoust. Soc. Am. Suppl. 1 82, S22 (1987)]. A small box centered at the mouth was extracted and spatially subsampled. The output of the neural net was the acoustic spectral envelope of the corresponding acoustic signal obtained via the cepstral method. After being trained on 95 tokens of vowels and dipthongs, the network was tested on 24 images it had not seen. The net was able to estimate the spectral envelope associated with these images more accurately than more traditional approaches that used stored templates. [Work supported by AFOSR Contract No. 86-0246.]

The acoustics of the vocal tract in the horseshoe bat, R h i n o l o p h u s h i l d e b r a n d t i

The transfer function of the supraglottal vocal tract of the horseshoe bat, Rhinolophus hildebrandti, was obtained by a noninvasive technique, based on incremental variations in the helium content of inspired gas, which also allowed the source spectrum to be determined. The acoustic role of the vocal tract chambers was examined by obtaining the transfer function before and after filling the chambers. Simultaneous recording of sound pressures in the trachea during these experiments allowed some analysis of subglottal acoustics. With the vocal tract intact, the transfer function was found to show sharp transmission minima at the fundamental and third harmonic and a broad transmission maximum at the emitted second harmonic. This transfer function shape, along with the source spectrum obtained, demonstrates that the second harmonic dominance in the emitted pulse is achieved by vocal tract filtering, although the source spectrum is different from that typical of man in that it does not show an f−2 harmonic decay. Changes in the transfer function caused by filling the nasal chambers suggest that these structures may play an impedance matching role at the second harmonic. Filling of the tracheal chambers did not affect the transfer function but changed the tracheal acoustics in a manner which suggests that these chambers may return backward-propagated sound to the larynx with a phase shift. The possible interactive role of the nasal and tracheal chambers in increasing vocal efficiency at the second harmonic is discussed.

Voiced speech model including source‐tract interaction

This paper considers the problem of estimating the parameters of a voiced speech model that includes the effects of source‐tract interaction. Based on work by Ananthapadmanabha and Fant [Speech Commun. 1, 167–184 (1982)], the voiced speech signal is modeled as the output of a time‐varying vocal tract filter excited by a parametrized glottal source waveform. The glottal source waveform [Fant et al., STL/QPSR (1984)] models the main pulse shape of the glottal volume velocity and is described by four parameters. The effect of source‐tract interaction is modeled by varying the frequency and bandwidth of the first formant in synchrony with the glottal source waveform. An analysis‐by‐synthesis approach is used to estimate the parameters of the model. Algorithms for estimating the parameters of the glottal source waveform and the vocal tract filter are described. Comparisons of the spectrum of the original and synthesized speech waveforms are presented. [Work supported by the OSU Seed grant program.]

Digital inverse filtering for linguistic research.

To enable differences in modes of glottal vibration to be studied, glottal air volume velocity waveforms can be recovered from speech recordings by inverse filtering. Most previous published work in this area has made use of analog filters. Digital inverse filters offer many advantages, including the ability to change filter settings to match changing vocal tract filter functions. Although the theory and many of the methods necessary for digital inverse filtering have been described in the literature, a straightforward description of the entire process has been lacking. The digital inverse filtering process developed for linguistic research at the UCLA Phonetics Laboratory is described in detail in this paper, with the intention of facilitating such work at other institutions.

Glottal flow: models and interaction

The present article summarizes recent work at the Royal Institute of Technology in Stockholm on glottal flow models and source-filter interaction. A four-parameter model of the glottal flow derivative (Fant, Liljencrants & Lin, 1985) has been introduced. It incorporates a return phase after the instant of maximum closure discontinuity, the effective duration of which is inversely related to a spectrum fall-off parameter. The interaction of supraglottal pressure build-up and glottal flow profile accounts for a dependency of voice source characteristics on the vocal-tract filter function (Fant, Lin & Gobl, 1985). The F1 component of supraglottal pressure conditions a substantial part of the interaction. It is demonstrated that a phonation with F1 close to F0 minimizes air consumption, as suggested by Rothenberg (1985). This is also to a less degree found when F1 comes close to 2F0. Other interaction phenomena discussed are the occasional occurrence of a double peak in the glottal flow derivative and perturbations in the wave shape and amplitude of successive glottal flow pulses. At a high pitch, it is the supraglottal pressure build-up from previous periods that dominates the interaction. Glottal friction does not contribute much to the flow waveform and its excitational power.

Automatic glottal inverse filtering from speech and electroglottographic signals

The flow of air through the glottis, the glottal volume-velocity, reflects the action of the vocal folds and is thus an important indicator of laryngeal function. However, it cannot be measured directly because of vocal tract filtering. We have developed an automated on-line method to determine the glottal volume-velocity waveform from normal and pathological speech based on digital inverse filtering. The method developed addresses the problems of accurate identification of vocal tract parameters and reduction of low-frequency noise. The vocal tract filter is estimated by analysis of the undriven vocal tract response during closed glottis, as identified from an electroglottographic signal. Low-frequency noise is attenuated by a high-pass filtering operation followed by a low-pass compensation. The complete inverse filtering method provides reliable glottal volume-velocity waveforms for both normal and pathological speech.

Speech recognition by polarized linear predictive error coding—POLPEC method

AbstractThis paper describes the polarized linear predictive error coding method (POLPEC), which is a new scheme for speech recognition. Although the residual waveform power method, which is a method of speech recognition using an inverse vocal tract filter, has the advantage that the processing for recognition is simple and is suited to parallel processing, it has the drawback that it is sensitive to deviation of the input waveform to the inverse filter (i.e., the speech waveform). From such a viewpoint, this paper introduces the polarized linear predictive error coding method, which is a recognition scheme, in which it is assumed that deviation in the speech waveform is due to change in the pole of the transfer function of the filter for speech generation (called vocal tract filter) and, by improving the residual waveform power method, the sensitivity to the change of poles is decreased. The POLPEC method has the effect of decreasing the residual power by subtracting from the residual waveform sequence the component which is due to the deviation of poles. It is verified by experiment using synthetic speech that this reduces the sensitivity to the predetermined deviation of poles.

Speech Enhancement Based on Linear Prediction and Correlation-Inputting Bias Free Equation Error ADF

In this paper, a speech enhancement technique to reduce background noise in noisy speech is proposed. We investigated the noise reconstruction system (NRS) based on linear prediction and system identification as a speech enhancement. Assuming that the background noise is generated from white noise by exciting a linear filter, the system identification estimates the background noise from estimated white noise. However, the white noise estimated by a linear prediction error filter (LPEF) includes residual speech, then the estimation accuracy of background noise is degraded at the system identification and the quality of enhanced speech is deteriorated. In order to reduce the influence of the residual speech, a lattice filter and a bias free equation error adaptive digital filter (ADF) are respectively introduced to the LPEF and system identification. The residual speech is reduced by the lattice filter which approximates a vocal-tract filter well. On the other hand, the bias free equation error ADF uses the cross-correlation between the whitened noise and a desired signal as a tap input. Since the speech does not have the correlation from the desired signal, the tap coefficients converge without the influence of speech.

The channel vocoder

The channel vocoder is described. This device achieves bandwidth compression greater than that of the bandpass compressor but less than that of the formant vocoder. To date it has been much more widely used than any other kind of vocoder. The channel vocoder exploits the insensitivity of the aural mechanism to phase, and only attempts to reproduce the short time power spectrum of the speech waveform. The spectral envelope of the speech is measured with a bank of filters and ascribed wholly to the vocal tract filter, while the excitation is estimated to be either a quasi-periodic pulse train, or noise. There are several methods of combining these extracted parameters to reconstruct the speech. Several configurations of the channel vocoder are described and the factors which affect the specification of design parameters for channel vocoders are considered.

Formant Tracking by Self-Adjusting Inverse Filtering

A method of formant tracking is described in which the speech wave is passed through an adjustable network whose transfer function approximates to the inverse of the vocal tract filter. From the output of this network it is possible to derive error signals which indicate the polarity of the unbalance for each formant frequency. These signals can be used for continuously improving the approximation so that the zeros in the inverse network continuously follow the poles, or formants, in the vocal tract filter. A test setup, using analog computer type elements, is described, and its performance examined both with synthetic input and with real speech input.