Voiced Speech Sounds Research Articles

Symptom Expression Across Voiced Speech Sounds in Adductor Laryngeal Dystonia

Differential diagnosis for adductor laryngeal dystonia (AdLD) is often carried out by comparing symptom expression during sentences with either all voiced or voiced and voiceless consonants. However, empirical research examining the effects of phonetic context on symptoms is sparse. The purpose of this study was to examine whether symptom probabilities varied across voiced speech segments in an all-voiced sentence, and whether this variability was systematic with respect to phonetic features. Eighteen speakers with AdLD read aloud a sentence comprised entirely of voiced speech sounds. Speech segment boundaries and AdLD symptoms (phonatory breaks, frequency shifts, and creak) were labeled separately, and speech segments were coded as symptomatic or asymptomatic based on their temporal overlap. Generalized linear mixed effects models with a binomial outcome variable were used to compare the probability of symptom expression across: 1) all speech segments in the sentence, and 2) four speech sound classes (vowels, approximants, nasals, and obstruents). Significant symptom variability was found across voiced speech segments in the sentence. Furthermore, the estimated probability of a symptom occurring on vowels and approximants was significantly greater than that of nasals and obstruents. These results indicate that AdLD symptoms are not uniformly distributed across voiced speech segments with systematic variation across speech sound classes.To explain these findings, future work should investigate how the complex interactions between the vocal tract articulators and glottal configurations may influence symptom expression in this population.

Spasmodic dysphonia: introductory phonetic analyses

ABSTRACT Adductor spasmodic dysphonia (ADSD) is a neurological dystonia characterised by involuntary adductor spasms of the larynx during speech. Spasm frequency is often reported to increase during syllables that begin with voiced speech sounds, especially glottal stops. Because of its underlying physical and acoustic complexities, the voicing contrast in American English (AE) appears unlikely to interact consistently with a singular physical phenomenon like laryngeal spasm. This retrospective study investigated additional phonetic contrasts and their relationship to spasm frequency. Standardised, 144-word recordings of 36 participants with adductor spasmodic dysphonia were analysed. Productions were coded for rater-perceived syllable stress, voiced/voiceless onset, vowel/consonant onset, and word-onset place and manner of production. Phonetic contexts were compared using independent sample t-tests and Kruskal-Wallis statistics. Contexts in which spasm varied significantly included stressed/unstressed syllables, content/function words, and multisyllabic/monosyllabic words. Study results reaffirm the clinical usefulness of standardised ADSD/ABSD sentences during differential diagnosis but conflict with previous studies that report a connection between ADSD spasm and phoneme voicing.

Modelling speaker-size discrimination with voiced and unvoiced speech sounds based on the effect of spectral lift

We can estimate the size of a speaker solely from their speech sounds, regardless of whether the sounds are voiced or unvoiced. In this study, we developed a size perception model based on the computational theory of the stabilised wavelet transform (SWT) to explain a variety of size discrimination data. We also conducted extended experiments to evaluate the effect of spectral lift on speaker size discrimination, from voiced and unvoiced speech sounds. The just noticeable difference (JND) and the point of subjective equality (PSE) for speaker size discrimination were compared between speech sounds with natural and lifted spectra. On average, listeners tended to judge that the lifted speech came from a smaller speaker. The PSE, which indicates the systematic difference in perceived size, shifted by approximately 10% (Exp. 1) for unvoiced speech sounds, and by approximately 5% (Exp. 2) for voiced speech sounds. The JND depended on the spectral lift for unvoiced sounds, but not with voiced sounds. At the same time, it was noted that there were large differences between listeners: some listeners’ judgements were affected by the spectral lift, while others were not. We constructed a size discrimination model to explain all of the experimental results with listener dependence for voiced and unvoiced speech sounds. We introduced a weighting function, based on the Size-Shape Image (SSI) in the SWT, which reduces the influence of resolved harmonics caused by the glottal pulse sequence in voiced speech. As a result, the model with the SSI weighting function predicted fairly well the individual listener’s data, whether the judgements were affected by the spectral lift or not, and whether the speech sounds were voiced or unvoiced. The optimum choice of one parameter, that is, the spectral compensation coefficient, enabled us to explain the data of all individuals.

Orangutans show active voicing through a membranophone

Active voicing – voluntary control over vocal fold oscillation – is essential for speech. Nonhuman great apes can learn new consonant- and vowel-like calls, but active voicing by our closest relatives has historically been the hardest evidence to concede to. To resolve this controversy, a diagnostic test for active voicing is reached here through the use of a membranophone: a musical instrument where a player’s voice flares a membrane’s vibration through oscillating air pressure. We gave the opportunity to use a membranophone to six orangutans (with no effective training), three of whom produced a priori novel (species-atypical) individual-specific vocalizations. After 11 and 34 min, two subjects were successful by producing their novel vocalizations into the instrument, hence, confirming active voicing. Beyond expectation, however, within <1 hour, both subjects found opposite strategies to significantly alter their voice duration and frequency to better activate the membranophone, further demonstrating plastic voice control as a result of experience with the instrument. Results highlight how individual differences in vocal proficiency between great apes may affect performance in experimental tests. Failing to adjust a test’s difficulty level to individuals’ vocal skill may lead to false negatives, which may have largely been the case in past studies now used as “textbook fact” for great ape “missing” vocal capacities. Results qualitatively differ from small changes that can be caused in innate monkey calls by intensive months-long conditional training. Our findings verify that active voicing beyond the typical range of the species’ repertoire, which in our species underpins the acquisition of new voiced speech sounds, is not uniquely human among great apes.

Midbrain sensitivity to frequency “chirps:” Implications for coding voiced speech sounds

Many neurons in the mammalian midbrain (inferior colliculus, IC) have strong selectivity for direction of frequency transitions (“chirps”), which often co-exist with the better-known sensitivity for amplitude-modulation (AM) frequency. Chirps occur when harmonics with a phase gradient are summed, as in Schroeder-phase harmonic stimuli. Chirps also occur in voiced speech due to the phase properties of vocal tract resonances. IC neurons are direction-selective for chirps in stimuli with fundamental frequencies in the voice-pitch range, and chirp selectivity can be achieved in an AM-tuned IC model by adding off-CF inhibition. Here, we will show physiological and model responses illustrating how chirps influence IC responses to speech in awake rabbit. In healthy auditory-nerve (AN) responses, the frequency extent, or size, of chirps vary with proximity of AN tuning to formant peaks due to inner ear nonlinearities. For example, synchrony capture causes one harmonic to dominate responses of AN fibers near spectral peaks, and responses that are dominated by one harmonic do not exhibit chirps. The systematic variation in chirp size across frequency channels and strong selectivity of IC neurons for chirps suggest a role for this feature in coding voiced speech. Importantly, this coding mechanism is vulnerable to sensorineural hearing loss.

Vocal similarity predicts the relative attraction of musical chords

Musical chords are combinations of two or more tones played together. While many different chords are used in music, some are heard as more attractive (consonant) than others. We have previously suggested that, for reasons of biological advantage, human tonal preferences can be understood in terms of the spectral similarity of tone combinations to harmonic human vocalizations. Using the chromatic scale, we tested this theory further by assessing the perceived consonance of all possible dyads, triads, and tetrads within a single octave. Our results show that the consonance of chords is predicted by their relative similarity to voiced speech sounds. These observations support the hypothesis that the relative attraction of musical tone combinations is due, at least in part, to the biological advantages that accrue from recognizing and responding to conspecific vocal stimuli.

Speech Coding in the Midbrain: Effects of Sensorineural Hearing Loss.

In response to voiced speech sounds, auditory-nerve (AN) fibres phase-lock to harmonics near best frequency (BF) and to the fundamental frequency (F0) of voiced sounds. Due to nonlinearities in the healthy ear, phase-locking in each frequency channel is dominated either by a single harmonic, for channels tuned near formants, or by F0, for channels between formants. The alternating dominance of these factors sets up a robust pattern of F0-synchronized rate across best frequency (BF). This profile of a temporally coded measure is transformed into a mean rate profile in the midbrain (inferior colliculus, IC), where neurons are sensitive to low-frequency fluctuations. In the impaired ear, the F0-synchronized rate profile is affected by several factors: Reduced synchrony capture decreases the dominance of a single harmonic near BF on the response. Elevated thresholds also reduce the effect of rate saturation, resulting in increased F0-synchrony. Wider peripheral tuning results in a wider-band envelope with reduced F0 amplitude. In general, sensorineural hearing loss reduces the contrast in AN F0-synchronized rates across BF. Computational models for AN and IC neurons illustrate how hearing loss would affect the F0-synchronized rate profiles set up in response to voiced speech sounds.

Reverberation impairs brainstem temporal representations of voiced vowel sounds: challenging "periodicity-tagged" segregation of competing speech in rooms.

The auditory system typically processes information from concurrently active sound sources (e.g., two voices speaking at once), in the presence of multiple delayed, attenuated and distorted sound-wave reflections (reverberation). Brainstem circuits help segregate these complex acoustic mixtures into “auditory objects.” Psychophysical studies demonstrate a strong interaction between reverberation and fundamental-frequency (F0) modulation, leading to impaired segregation of competing vowels when segregation is on the basis of F0 differences. Neurophysiological studies of complex-sound segregation have concentrated on sounds with steady F0s, in anechoic environments. However, F0 modulation and reverberation are quasi-ubiquitous. We examine the ability of 129 single units in the ventral cochlear nucleus (VCN) of the anesthetized guinea pig to segregate the concurrent synthetic vowel sounds /a/ and /i/, based on temporal discharge patterns under closed-field conditions. We address the effects of added real-room reverberation, F0 modulation, and the interaction of these two factors, on brainstem neural segregation of voiced speech sounds. A firing-rate representation of single-vowels' spectral envelopes is robust to the combination of F0 modulation and reverberation: local firing-rate maxima and minima across the tonotopic array code vowel-formant structure. However, single-vowel F0-related periodicity information in shuffled inter-spike interval distributions is significantly degraded in the combined presence of reverberation and F0 modulation. Hence, segregation of double-vowels' spectral energy into two streams (corresponding to the two vowels), on the basis of temporal discharge patterns, is impaired by reverberation; specifically when F0 is modulated. All unit types (primary-like, chopper, onset) are similarly affected. These results offer neurophysiological insights to perceptual organization of complex acoustic scenes under realistically challenging listening conditions.

Expression of Emotion in Eastern and Western Music Mirrors Vocalization

In Western music, the major mode is typically used to convey excited, happy, bright or martial emotions, whereas the minor mode typically conveys subdued, sad or dark emotions. Recent studies indicate that the differences between these modes parallel differences between the prosodic and spectral characteristics of voiced speech sounds uttered in corresponding emotional states. Here we ask whether tonality and emotion are similarly linked in an Eastern musical tradition. The results show that the tonal relationships used to express positive/excited and negative/subdued emotions in classical South Indian music are much the same as those used in Western music. Moreover, tonal variations in the prosody of English and Tamil speech uttered in different emotional states are parallel to the tonal trends in music. These results are consistent with the hypothesis that the association between musical tonality and emotion is based on universal vocal characteristics of different affective states.

Effects of Age on F0 Discrimination and Intonation Perception in Simulated Electric and Electroacoustic Hearing

Recent research suggests that older listeners may have difficulty processing information related to the fundamental frequency (F0) of voiced speech. In this study, the focus was on the mechanisms that may underlie this reduced ability. We examined whether increased age resulted in decreased ability to perceive F0 using fine-structure cues provided by the harmonic structure of voiced speech sounds or cues provided by high-rate envelope fluctuations (periodicity). Younger listeners with normal hearing and older listeners with normal to near-normal hearing completed two tasks of F0 perception. In the first task (steady state F0), the fundamental frequency difference limen (F0DL) was measured adaptively for synthetic vowel stimuli. In the second task (time-varying F0), listeners relied on variations in F0 to judge intonation of synthetic diphthongs. For both tasks, three processing conditions were created: eight-channel vocoding that preserved periodicity cues to F0; a simulated electroacoustic stimulation condition, which consisted of high-frequency vocoder processing combined with a low-pass-filtered portion, and offered both periodicity and fine-structure cues to F0; and an unprocessed condition. F0 difference limens for steady state vowel sounds and the ability to discern rising and falling intonations were significantly worse in the older subjects compared with the younger subjects. For both older and younger listeners, scores were lowest for the vocoded condition, and there was no difference in scores between the unprocessed and electroacoustic simulation conditions. Older listeners had difficulty using periodicity cues to obtain information related to talker fundamental frequency. However, performance was improved by combining periodicity cues with (low frequency) acoustic information, and that strategy should be considered in individuals who are appropriate candidates for such processing. For cochlear implant candidates, this effect might be achieved by partial electrode insertion providing acoustic stimulation in the low frequencies or by the combination of a traditional implant in one ear and a hearing aid in the opposite ear.

A map of periodicity orthogonal to frequency representation in the cat auditory cortex.

Harmonic sounds, such as voiced speech sounds and many animal communication signals, are characterized by a pitch related to the periodicity of their envelopes. While frequency information is extracted by mechanical filtering of the cochlea, periodicity information is analyzed by temporal filter mechanisms in the brainstem. In the mammalian auditory midbrain envelope periodicity is represented in maps orthogonal to the representation of sound frequency. However, how periodicity is represented across the cortical surface of primary auditory cortex (AI) remains controversial. Using optical recording of intrinsic signals, we here demonstrate that a periodicity map exists in primary AI of the cat. While pure tone stimulation confirmed the well-known frequency gradient along the rostro-caudal axis of AI, stimulation with harmonic sounds revealed segregated bands of activation, indicating spatially localized preferences to specific periodicities along a dorso-ventral axis, nearly orthogonal to the tonotopic gradient. Analysis of the response locations revealed an average gradient of − 100° ± 10° for the periodotopic, and −12° ± 18° for the tonotopic map resulting in a mean angle difference of 88°. The gradients were 0.65 ± 0.08 mm/octave for periodotopy and 1.07 ± 0.16 mm/octave for tonotopy indicating that more cortical territory is devoted to the representation of an octave along the tonotopic than along the periodotopic gradient. Our results suggest that the fundamental importance of pitch, as evident in human perception, is also reflected in the layout of cortical maps and that the orthogonal spatial organization of frequency and periodicity might be a more general cortical organization principle.

Low-frequency vocal modulations in vowels produced by Parkinsonian subjects

Low-frequency vocal modulations here designate slow disturbances of the phonatory frequency F 0. They are present in all voiced speech sounds, but their properties may be affected by neurological disease. An analysis method, based on continuous wavelet transforms, is proposed to extract the phonatory frequency trace and low-frequency vocal modulation in sustained speech sounds. The method is used to analyze a corpus of vowels uttered by male and female speakers, some of whom are healthy and some of whom suffer from Parkinson’s disease. The latter present general speech problems but their voice is not perceived as tremulous. The objective is to discover differences between speaker groups in F 0 low-frequency modulations. Results show that Parkinson’s disease has different effects on the voice of male and female speakers. The average phonatory frequency is significantly higher for male Parkinsonian speakers. The modulation amplitude is significantly higher for female Parkinsonian speakers. The modulation frequency is significantly higher and the ratio between the modulation energies in the frequency bands [3 Hz, 7 Hz] and [7 Hz, 15 Hz] is significantly lower for Parkinsonian speakers of both genders.

System for measuring velar function during speech

A method of and device for the diagnosis and treatment of speech dynamically measures the functioning of the velum in the control of nasality during speech. Various components of oral and nasal airflow are separated and selectively analyzed including (i) the fundamental frequency component of each airflow during voiced speech, (ii) a plurality of voice components that cover a frequency range encompassing at least the lowest vocal tract resonance (the first formant), and (iii) the subsonic and infrasonic components of at least the nasal airflow. By comparing the nasal and oral airflow components at the voice fundamental frequency, a nasalization measure for voiced speech sounds is formed which emulates methods that compare low frequency nasal and oral airflow during voiced speech, while eliminating or greatly reducing the problems associated with comparing these low frequency airflows, and which improves upon previous methods based on measuring and comparing nasal and oral radiated sound pressure. A circumferentially vented screen mask (C-V mask) is configured with separate nasal and oral chambers to separate the two airflows, and causes only a minimal distortion and muffling of the voice. The separate nasal and oral airflows are detected and filtered, and a ratio of the two is formed to provide a visual display used to detect and correct abnormal or incorrect speech formation and word pronunciation.

Aeroacoustic mechanisms of voiced sound production

The focus of this study is to quantify the order of magnitude of the direct effects of (1) vocal-fold wall motion and (2) glottal flow separation point movement on the production of voiced speech sounds. A solution for the sound-pressure field shows three source mechanisms: (1) a volume source due to unsteady glottal air flow; (2) a quadrupole source representing interaction of the glottal jet with the pharynx walls; and (3) an octupole due to direct sound radiation by the glottal jet itself. A relation is derived expressing glottal volume flow in terms of transglottal pressure difference, vocal-fold wall motion, and separation point motion. Using scaling analysis, the transglottal pressure difference is shown to be the dominant effect on glottal volume flow, while vocal-fold wall motion is shown to have a negligible effect. However, separation point motion is shown to have a measurable effect during the closure phase of the vibration cycle. Using these results, the acoustic effect of separation point motion is shown to be measurable, while the effect of vocal-fold wall vibration is shown to be negligible. Relative contributions of these effects across age, gender, and degree of glottal closure are discussed.

Stochastic models of jitter.

This study presents stochastic models of jitter. Jitter designates small, random, involuntary perturbations of the glottal cycle lengths. Jitter is a base-line phenomenon that may be observed in all voiced speech sounds. Knowledge of its properties is therefore relevant to the acoustic modeling, analysis, and synthesis of voice quality. Also, models of jitter are conceptual frameworks that enable experimenters and clinicians to distinguish jitter in particular from aperiodic cycle length patterns in general. Vocal jitter is modeled by means of the ribbon model of the glottal vibration combined with stochastic models of the disturbances of the instantaneous frequency. The disturbance model comprises correlation-free noise and vocal microtremor. Properties of jitter that are simulated are the stochasticity, stationarity, and normality of the decorrelated cycle length perturbations, the size of decorrelated jitter, the correlation between the perturbations of neighboring glottal cycles, the modulation level and modulation frequency owing to microtremor, the asynchrony between external disturbances and glottal cycles, the dependence of the size of jitter on the average glottal cycle length, and the relation between jitter and laryngeal pathologies. Modeled jitter is discussed in the light of measured jitter, as well as the physiological and statistical plausibility of the model parameters.

The integration of discriminative and maximum likelihood distance measures for a vowel articulation training aid

Improvements to the vowel articulation training aid described in a previous paper [Zimmer, Zahorian, and Auberg, J. Acoust. Soc. Am. 101, 3199(A) (1997)] have been made. The system uses a standard Windows 95/NT compatible sound card on a multimedia PC to provide continuous feedback about articulation for ten American English monopthong vowels in two modes: an F1/F2 style ‘‘ellipse’’ display and a vowel bargraph display. Neural network discriminative classifiers are used to produce the display outputs based on 12 features. Though testing showed that the system provided useful output for most voiced speech sounds, the network would classify some out-of-category sounds (e.g., nonvowel sounds, ambient noise) as belonging to one of the ten vowel categories. Experimental tests indicate that the inclusion of a generalized Euclidean distance measure to compare the feature values of an utterance with the average feature values for the vowel category specified by the neural network output help to greatly reduce the number of out-of-category sounds improperly classified as ‘‘correct’’ responses. The paper will describe the processing in more detail and summarize experimental results from vowel classification tests. [Funded by NSF, Grant No. NSF-BES-9411607.]

Pure-tone increment detection in harmonic and inharmonic backgrounds

The present study investigated the effect of increasing the number and frequency range of background components upon the detection of an increment to the intensity of the center component. Previous studies [e.g., Green et al., J. Acoust. Soc. Am. 75, 1163-1167 (1984)] have demonstrated that, for background stimuli consisting of logarithmically spaced pure-tone components, as the number of background components is increased, the threshold for detection of an increment to the center component is decreased. The present study investigated this effect using harmonically spaced, as well as logarithmically spaced, backgrounds, because many natural sounds, including voiced speech sounds, consist of harmonically spaced components. Two conditions were employed for both types of backgrounds, fixed-level and roving-level within a trial. Group results revealed a small decrease in threshold with increasing number of components only for the fixed-level, logarithmically spaced backgrounds. No decrease in threshold was observed for the other three conditions, including the roving-level, logarithmically spaced backgrounds which were the same as those used by Green et al. (1984). The present results suggest that the decrease in signal threshold with increasing number of background components exists only under limited conditions and that it is a highly individual phenomenon. The present results also suggest that the effect does not occur for harmonically spaced complexes such as most natural sounds including voiced speech.

On the measurement of glottal flow.

For developing a comprehensive description of voiced speech sounds in terms of a phonation and an articulation component, it is necessary to know to what extent the volume flow modulations at the entrance of the vocal tract are due to vocal fold motions and to what extent they are due to variations in the transglottal pressure. In order to be able to study this problem, it is important that the flow at the glottis can be measured during normal speech production in a reliable fashion. In this article, a flow measurement technique is described that differs from the more usual inverse filtering approach to the extent that the flow is not measured at the mouth, but much closer to the glottis. The technique is based on the measurement of pressure gradient. It is shown that the proposed method also leads to an inverse filtering problem, but that, since this problem is much simpler, the gradient method yields more reliable estimates of the shape of the glottal flow waveform, though without the zero flow level (dc component) and without a magnitude scale. By means of theoretical considerations about velocity profiles in pulsatile flow in cylindrical tubes, it is shown that the method for measuring flow during phonation proposed in this article may be expected to yield reasonable flow waveform estimates in a frequency region from any normal fundamental frequency to an upper frequency determined by the transducer sensitivity and separation and vocal tract geometry. In this case, the frequency limitation was estimated to be 1000 Hz.(ABSTRACT TRUNCATED AT 250 WORDS)

Formant estimation algorithm based on pole focusing offering improved noise tolerance and feature resolution

The ability to measure the centre frequencies of areas of resonance (formants) in the short-time power spectrum of speech is of paramount importance in the recognition of voiced speech sounds in a feature-extraction-based continuous speech recognition system. Additionally, the provision of a tracking algorithm, by which the loci of formants with respect to time can be estimated, yields formant transition information which helps identify phonetic features which are of short duration. Noise robustness in formant estimation is an essential attribute for recognition systems which are used in the office environment and in military applications. The novel technique presented in the paper provides a noise-robust method of extracting formant centre-frequency information from the short-time speech spectrum, and consequently improves the signal/noise performance of the associated formant tracking algorithm. Formant estimation is based on modelling the vocal tract frequency response using linear prediction coding (LPC) techniques. However, the estimation of formant centre frequency in any given analysis frame is greatly improved by employing off-axis spectral estimation coupled with a progressive increase in vocal tract model order, which together provide vocal tract pole enhancement. Finally, the use of a formant weighting filter function applied within each frame aids in conferring high noise immunity to the estimation process. The pole focusing technique is shown to offer an improvement of at least 14 dB in signal/noise immunity as a formant frequency estimator over conventional LPC-based spectral estimation. In its application to formant tracking, it is shown that the technique also offers improved separation of formants which tend to merge, besides offering a general improvement in the provision of formant detail, in particular with regard to weak nasal formants. An additional advantage of the technique is its relative insensitivity to choice of vocal tract model order, which produces an inherently speaker-independent formant estimation algorithm.

The time-domain periodogram algorithm

The Time-Domain Periodogram Algorithm (TDPA) is proposed for estimating the pitch period of voiced speech sounds. After the algorithm has been presented theoretically, experimental results of the analysis of speech signals and sinusoids using the TDPA are given and the algorithm's noise performance is also evaluated. These results compare favourably with the average magnitude difference function (AMDF). The TDPA is highly suitable for implementation on a 16-bit microprocessor, since it uses no multiply operations and the computation can be performed in integer arithmetic without exceeding the dynamic range of the microprocessor. The TDPA also provides as a byproduct a well behaved estimate of signal intensity.