Voice Quality Variation Research Articles

Automatic creation of CV templates for formant type speech synthesis based on HMM-based segmentation and syllable boundary detection

A formant type speech synthesis method has an enormous advantage in that it allows one to generate speech with various voice quality variations and talker individualities. But it has suffered from unnatural speech quality, not because of theoretical limitations but because of an incomplete set of rules for the synthesis. Any insufficient approximation to the acoustics causes degradation of the perceived quality of synthetic speech. A novel formant type speech synthesizer in Japanese based on concatenation of CV (consonant-vowel) formant-source templates obtained from natural utterances has been investigated, in which multiple sets of formant and voice source parameter values are used for each of the CV syllables. This paper describes an automatic method to create the CV formant-source templates from speech corpus. The ARX (autoregressive with exogenous input) analysis method is first used to automatically extract formant and voice source parameters and then an HMM based segmentation is performed to locate the CV segments. The segments are further analyzed to detect a starting point of the syllable. A distance measure is used to decide the number of templates needed for each of the syllables. The method is proved to be useful in creating CV templates by experiments performed on 503 Japanese sentences.

Acoustic characteristics of voice quality

Subtle variations in voice (phonatory) quality may reveal aspects of the speaker's mood and attitude, and are thus an important aspect of speaking style. This paper illustrates current research being carried out by the authors on the voice source correlates of a range of such quality differences. The voice qualities looked at include modal, breathy, whispery, tense, lax and creaky voice, following the description of Laver (1980) and analyses presented here focus on a word extracted from a propose passage read with these qualities. The principal method used for analysing the voice source involved inverse filtering of the speech wave. In order to quantify the source characteristics, a four parameter voice source model (the LF-model) was matched to the inverse filtered waveform. Frequency domain analyses of the speech waveform, based on narrow band spectral sections and on spectral averaging, were also carried out. Detailed comparisons of the data measured directly from the glottal waveform and those measured from the speech output yield insights which could not be inferred from either alone. Results suggest a number of important differences between the qualities as well as considerable dynamic variation within a single quality. The data should also prove useful for resynthesis, which is an important tool for testing perceptual aspects of voice quality; e.g., the attitudinal and emotional colouring that may be associated with particular voice qualities.

A Technique for Analysing Voice Quality in Pathological and Normal Speech

A technique for voice source analysis is described which is currently being used to study voice quality variation, in dysphonic and normal speakers. The analysis technique involves a two-step procedure. The first, inverse filtering of the speech pressure waveform, yields the differential glottal flow. The settings for the inverse filter are initially estimated using a pitch-synchronous semi-automatic procedure based on the covariance method of LPC. The filter is then interactively edited on a pulse-by-pulse basis to correct for errors to which current automatic procedures are prone. The second step in the analysis involves matching a voice source model (the Liljencrants-Fant model) to the output of the inverse filter, again on a pulse-by-pulse basis. This allows quantification and comparison of the salient voice source characteristics. An illustration of some voice quality variation in normal and dysphonic speech is presented.

Spectral parameters of a voice source pulse.

This study reports systematic covariations between the time and frequency domain parameters of a voice source pulse. Open quotient, speed quotient, rate of closure, and the maximum negative amplitude of the voice source pulse are the time domain parameters. Frequency of the spectral peak, spectral roll-off, spectral level at fundamental frequency (F0), and twice F0 are some of the spectral parameters. A normalized representation for spectral parameters is proposed. Analysis methods to measure these normalized parameters reliably from the inverse-filtered signal are presented. The methods presented in this paper are generally useful in natural sounding speech synthesis and for studying variations in voice quality. Illustrative examples are presented of voice source variation with respect to phonetic and linguistic context and for different voice qualities. The methods may also be used in evaluating the voice quality of coded speech and in the diagnosis of voice disorders.

Objective acoustic voice-quality parameters in the computer phonetogram

Summary A phonetogram is a plot of the dynamic range of the voice as a function of fundamental frequency. Traditionally, the phonetogram only records the sound-pressure level (SPL) of the threshold of phonation and the upper limit in SPL the voice can reach with sustained phonation as a function of the fundamental frequency ( F 0 ). In this study, the phonetogram is extended by including acoustic voice-quality parameters. Three additional parameters are tested: jitter, shimmer, and crest factor. For each individual voice, the variation in the three parameters is evaluated over the entire phonetogram area. By averaging individual phonetograms, standard or norm phonetograms are obtained revealing general patterns in voice-quality variation. These patterns reflect the complex relations between F 0 , SPL, and the acoustic voice-quality parameters just mentioned. Jitter and shimmer distributions over the phonetogram show that large variations in perturbation values can be expected when production conditions are altered. Highest perturbation values are found for the low F 0 and low SPL phonations. For all voices, a gradual decrease of the crest factor is found with increasing F 0 , reflecting the degrading of spectral complexity with F 0 . The crest-factor parameter can also be used to mark those areas in the phonetogram where the fundamental dominates the waveform and where flow control is the main SPL regulating mechanism in voice production. The strong quality variations within the phonetogram stress the need for accurate control of F 0 and SPL in objective voice-quality measurement.

Toward a voice source characterization of voice quality

Voice source characteristics for voice quality variations in an individual nonpathological speaker are presented. This forms part of an ongoing analysis that is being carried out within the framework of Laver's classification [J. Laver, The Phonetic Description of Voice Quality (Cambridge U. P., Cambridge, 1980)] and is based on multiple repetitions by a trained phonetician of an English prose passage and nonsense materials, with modal (neutral), creaky, breathy, whispery, tense, and lax voice qualities. The objective is to provide an acoustic characterization of different “cardinal” qualities in the terms of both time and frequency domain parameters. Time domain measures are based on inverse filtered data to which a model of glottal flow (the LF model) has been matched. Frequency domain measures are based on spectral sections of the speech waveform and of the glottal waveform. As an initial strategy, templates of the different voice qualities are presented, based on the stressed vowel [æ]. However, data are also presented to demonstrate the considerable modulation of source quality even within what would traditionally be considered a single voice quality. For example, in creaky voice, the degree of creakiness is not uniform, but tends to be most extreme in unstressed syllables and at specific locations related to the prosodic profile of the utterance. Thus the differences between qualities may be more or less great depending on the environment compared. [Work supported by ACCOR, Esprit, BRA, No. 3279, and the Swedish Institute.]

Analysis, synthesis, and perception of voice quality variations among female and male talkers

Voice quality variations include a set of voicing sound source modifications ranging from laryngealized to normal to breathy phonation. Analysis of reiterant imitations of two sentences by ten female and six male talkers has shown that the potential acoustic cues to this type of voice quality variation include: (1) increases to the relative amplitude of the fundamental frequency component as open quotient increases; (2) increases to the amount of aspiration noise that replaces higher frequency harmonics as the arytenoids become more separated; (3) increases to lower formant bandwidths; and (4) introduction of extra pole zeros in the vocal-tract transfer function associated with tracheal coupling. Perceptual validation of the relative importance of these cues for signaling a breathy voice quality has been accomplished using a new voicing source model for synthesis of more natural male and female voices. The new formant synthesizer, KLSYN88, is fully documented here. Results of the perception study indicate that, contrary to previous research which emphasizes the importance of increased amplitude of the fundamental component, aspiration noise is perceptually most important. Without its presence, increases to the fundamental component may induce the sensation of nasality in a high-pitched voice. Further results of the acoustic analysis include the observations that: (1) over the course of a sentence, the acoustic manifestations of breathiness vary considerably--tending to increase for unstressed syllables, in utterance-final syllables, and at the margins of voiceless consonants; (2) on average, females are more breathy than males, but there are very large differences between subjects within each gender; (3) many utterances appear to end in a "breathy-laryngealized" type of vibration; and (4) diplophonic irregularities in the timing of glottal periods occur frequently, especially at the end of an utterance. Diplophonia and other deviations from perfect periodicity may be important aspects of naturalness in synthesis.

Variations in measured voice quality related to linguistic and nonlinguistic factors

Voice quality (VQ) in English has generally been treated as a suprasegmental, not varying across individual speech segments. Ladefoged [in Vocal Fold Physiology, edited by Bless and Abbs (1983)] has measured the spectral correlates of VQ differences in languages in which both VQ and tone (F0) are phonemically contrastive. Since VQ variation is not contrastive in English, it is in theory available to enhance other linguistic contrasts, or may vary as a function of nonlinguistic factors, such as fatigue or affect. Following Ladefoged, VQ was measured as the difference between the amplitude of F0 and the amplitude of the most intense harmonic in F1 using discrete Fourier transform spectra. Two separate corpora of utterances by subjects without known voice pathology were analyzed. One corpus compared qualitatively similar stressed and unstressed vowels (Vs) produced by a single speaker in “fresh” and “fatigued” circumstances. The other compared utterances produced by two trained actors in contrastively and normally stressed conditions, and simulated positive and negative effective conditions. These analyses suggest that this measure of VQ differentiates between Vs in (1) stressed and unstressed conditions, and (2) normally and contrastively stressed conditions. However, it did not differentiate between Vs in fresh and fatigued conditions or in the simulated affective conditions. The results are discussed in terms of the use of VQ and its relation to F0 for conveying linguistic and nonlinguistic distinctions. [Work supported by NIH Grants NS 07237, HD 1994, NS 24655.]

Speech changes in Parkinson's disease during treatment with L-DOPA

The speech of 17 Parkinsonian patients was evaluated before and after the administration of L-DOPA therapy. A significant difference was demonstrated after treatment for voice quality, articulation, and pitch variation, but not for rate of speech. Amount of speech improvement correlated significantly with amount of physical improvement. Although age and duration of disease may exert some influence on speech change, the results suggest that these factors do not reliably predict the response. After 4 years of L-DOPA therapy, three out of four patients demonstrated additional improvement or the same degree of speech improvement.

Variations in voice quality with laryngeal tumours and radiotherapy of the larynx.

Vocal characteristics of a number of patients who had radiotherapy for laryngeal tumours were obtained. It was found that voice quality before treatment was not indicative of the site and stage of the disease. Voice quality after treatment was usually fairly good.A formula is given for quantitative estimates of hoarseness and is followed by a brief discussion on some of its physical causes.