Synthetic Vowels Research Articles

Neuromagnetic responses elicited by synthetic vowel stimuli

Neuromagnetic responses elicited by synthetic vowel stimuli

The perception of complex harmonic patterns by profoundly hearing-impaired listeners.

In providing profoundly hearing-impaired persons with processed speech through a signal-processing hearing aid, it is important that the new speech code matches their auditory capacities. This processing capacity for auditory information was investigated in this study. In part 1, the subjects' ability to judge similarities among 8 different but related harmonic complexes was studied. The patterns contained different numbers of harmonics to a 125-Hz fundamental frequency; the harmonics had been spread over the spectrum in various ways. The perceptual judgments appeared to be based on a temporal cue, beat strength, and a spectral cue, related to the balance of high and low frequency components. In part 2, three sets of synthetic vowels were presented to the subjects. Each vowel was realized by summing harmonically related in-phase sinusoids at two formant frequencies. The sets differed in the number of sinusoids per formant: 1, 2 or 3. It was found that the subjects used spectral cues and vowel length for differentiating among the vowels. The overall results show the limited but perhaps usable ability of the profoundly impaired ear to handle spectral information. Implications of these results for the development of signal-processing hearing aids for the profoundly hearing impaired are discussed.

Masking of tones and formant frequencies

The purpose of this study was to find out whether results of auditory masking theory, established primarily for pure tones, can be used to predict masking of vowel formant frequencies by broadband noise maskers. For this purpose, three sets of 2AFC experiments were conducted. In each experiment, four frozen white-noise waveforms (with similar statistical properties) were used. The task in the first set of experiments, was tone-in-noise detection in which the tones were located at 1100 and 1800 Hz. In the second set, one-formant synthetic vowels were used in vowel-in-noise detection tasks. The formant was located at 1100 Hz in one experiment, and at 1800 Hz, in another. Finally, listeners were presented with pairs of four-formant synthetic vowels in noise: one pair had identical vowels and the second vowel in the second pair had a missing second formant frequency (F2). Subjects were instructed to choose the interval thought to contain the different pair of vowels. The four-formant synthetic vowel was either /ɑ /-like (formants at 625, 1100, 2700, and 3250 Hz) or /schwa/-like (formants at 500, 1800, 2700, and 3500 Hz). It was found that, on the average, the masked thresholds of tones and of formant frequencies were predicted well from the signal-to-noise ratio in a critical band centered around the tone or the formant frequency. [Work supported by NIH and NSF.]

A model of concurrent vowel segregation using an implementation of the subtraction strategy

The perceptual experiments performed by Lea [‘‘Auditory Modeling of Vowel Perception,’’ unpublished doctoral thesis, University of Nottingham (1992)] required listeners to identify the members of pairs of steady-state synthetic vowels presented monaurally, called concurrent vowels. Accuracy of identification was higher when one vowel was voiced and the other whispered compared to control conditions in which both were whispered or both were voiced with the same fundamental frequency (f0). Surprisingly, the improvement in accuracy was restricted to the whispered member of the voiced/whispered pair. This outcome is compatible with the idea that one strategy used by listeners to recover a target voice from a mixture of voices is to cancel an interfering voice by a process of spectrotemporal subtraction. Here, a computational model is described that implements this strategy. The model is based on a bank of bandpass filters and an array of autocorrelators. The period of the dominant pitch in a stimulus is estimated from the largest peak in the normalized sum of the autocorrelation functions. This estimate guides the synthesis of an array of autocorrelation functions that are subtracted from the array generated by the stimulus, hence, canceling an interfering voice. Following Meddis and Hewitt [J. Acoust. Soc. Am. 91, 233–245 (1992)], vowel identification is performed by matching the short-time part of the resulting summed autocorrelation functions to a set of stored templates. The model makes good qualitative predictions of the accuracy of listeners’ identification responses to the stimuli used in the perceptual experiments. In addition, it makes good quantitative predictions of the effects of introducing a difference in f0 between concurrent vowels reported previously by Assmann and Summerfield [J. Acoust. Soc. Am. 88, 680–697 (1990)]. [Work supported by the MRC of the UK.]

Using the method of adjustment to study vowel spaces.

Comparing the acoustic vowel spaces of different dialects or languages is fraught with peril because acoustic measurements confound personal and linguistic information. The study to be reported addressed this problem by using the method of adjustment. Listeners were asked to adjust the F1 and F2 of synthetic vowels to produce a perceptual match between the synthetic vowel and the vowel sound they would expect to hear in a visually presented word. The monosyllabic test words represented ten or eleven different vowel qualities (depending on dialect) for American English speakers. The listeners adjusted F1 and F2 by moving a mouse cursor, clicking the mouse button to hear the vowel sound indicated by the location of the cursor on a CRT. F3, F4, and the formant bandwidths were estimated by regression from previously reported acoustic data. Preliminary results suggested that: (1) Speakers of the same dialect were very consistent with each other; within subject variability (which was small) accounted for about 70% of all within vowel variability. (2) The method revealed perceptual vowel space differences between speakers of different dialects of American English. (3) Differences between monolingual and bilingual speakers of English from the same language community were also found. [Work supported by NIH.]

Preliminary experiences in using automatic inverse filtering of acoustical signals for the voice source analysis

A new straightforward glottal wave analysis method, called Iterative Adaptive Inverse Filtering (IAIF), is presented. The method is based on the separated speech production model that consists of three processes: The glottal excitation, the vocal tract and the lip radiation effect. In order to estimate the vocal tract transfer function the effect of the source to the speech spectrum is first estimated with an iterative procedure. After eliminating the estimated glottal contribution, the vocal tract transfer function is estimated using linear predictive analysis (LPC). The estimate for the glottal excitation is finally obtained by cancelling the effects of the vocal tract and lip radiation by inverse filtering. The performance of the new algorithm is verified and compared to two other LPC-based glottal wave analysis methods using synthetic vowels. It is shown that the IAIF-method is able to give a fairly accurate estimate for the glottal excitation. The results of the analysis of natural speech using the I...

Binaural masking level difference effects in single units of the guinea pig inferior colliculus

We have studied the masking effects of a binaurally presented noise on the responses to binaural signals recorded from low-frequency cells in the inferior colliculus of the guinea pig. The spike rates to the masker and signal + masker were compared to quantify masking at different interaural time delays of the noise. The signal was a 50-ms tone burst at best frequency or a 50-ms segment of a synthetic vowel, presented at the best interaural delay of the unit tested. At each noise masker delay, the noise level was adjusted to obtain a criterion spike difference. In most cases, the level required was lowest at the best delay for the noise. The mean difference between maximum and minimum masked thresholds across the cell population was very similar to the human psychophysical masking level difference under the same signal and masker conditions. In another series of tests, we measured the effect of the noise masker on the temporal pattern of the discharge to the signal. The signal used was a 500-ms segment of the synthetic vowel. In virtually all cases the addition of a continuous noise masker reduced the discharge rate synchronized to the fundamental frequency of the vowel. The degree of this reduction was dependent on the interaural time delay of the noise masker. For most units, maximum reduction was seen when the vowel and noise had the same interaural time delay. The similarity between the masking which we have shown physiologically and that reported in a variety of human psychophysical experiments suggests that the processing at levels up to and including the inferior colliculus contributes to the psychophysical BMLD.

Differential sensitivity to vowel continua in Old World monkeys (Macaca) and humans

Previous studies indicate that monkey pure tone frequency discrimination is quantitatively and qualitatively very different from that of humans: Monkey DLs at 1.0 and 2.0 kHz are up to 20 times larger than human DLs, and monkeys DLs increase as sensation level increases, in contrast to human DLs [Sinnott et al., J. Acoust. Soc. Am. 78, 1977-1985 (1985); Sinnott et al., J. Comp. Psychol. 101, 126-131 (1987)]. These results led to an hypothesis that monkey frequency discrimination is more dependent upon "rate" coding than is that of humans. The present study compared monkey and human DLs for formant frequency changes along three synthetic vowel continua /I-i/, /ae-epsilon/, and /a-v/. Here, monkey DLs for formants near 1.0 and 2.0 kHz (32-48 Hz) were only about two to three times larger than human DLs (11-21 Hz), and both monkeys and humans exhibited relatively similar, flat sensation level functions. Taken together, these data indicate that monkey and human frequency discrimination is more similar in the case of a complex vowel stimulus than in the case of a simple pure tone stimulus. Results are discussed in relation to "rate" versus "temporal" coding of tones and vowels in the auditory system.

Interaural delay sensitivity to tones and broad band signals in the guinea-pig inferior colliculus

We have measured the sensitivity of 243 low-frequency cells in the central nucleus of the guinea pig to the interaural time delay of best frequency (BF) tones, wideband noise and synthetic vowels. The highest rate of firing for the majority of cells occurred when the stimulus to the contralateral ear arrived 100–400 μs before that to the ipsilateral ear. The best delays for tones and noise measured in the same cell were highly correlated. In contrast to the tone delay functions, the majority of the delay functions obtained in response to wideband signals did not cycle, but were characterized by a single dominant peak or trough. The response frequency calculated from the delay functions to the vowel often did not correspond to the unit's BF, suggesting that the unit was responding to a component close to the first formant frequency (730 Hz) of the vowel. Phase-locked responses, on the other hand, only occurred to the fundamental frequency of the vowel (100 Hz) and not to higher frequency components. The responses to delayed tone and noise signals in the guinea pig are very like those obtained in the cat and other mammals. The similarity of the range of best delays for the guinea-pig with those reported for the cat, despite the difference in head size in these two species, suggests that the sensitivity to interaural delays reflects the properties of the binaural pathways rather than an adaptation to the delays normally experienced by the animal.

Pitch, vowel quality, and the perceptual segregation of competing voices

An important problem faced by models of the auditory and phonetic analysis of speech is describing how listeners extract information from speech when competing sound sources are present. When the competing source is another voice, the task for the auditory system is complicated by the high degree of similarity in the spectro‐temporal structure of the two sounds. This situation imposes special constraints on models of auditory processing. Some of these constraints are revealed in recent studies of the role of fundamental frequency (f0) and formant structure for grouping and segregating the acoustic components of two competing voices. In one set of experiments, listeners attend to “double vowels” created by mixing the waveforms of pairs of simultaneous synthetic vowels with similar overall amplitudes. Both constituents of double vowels can be identified with an accuracy greater than chance, even when they have the same f0. However, when a difference in f0 is introduced between the constituents, listeners sh...

The representation of the spectra and fundamental frequencies of steady-state single- and double-vowel sounds in the temporal discharge patterns of guinea pig cochlear-nerve fibers

Psychophysical results using double vowels imply that subjects are able to use the temporal aspects of neural discharge patterns. To investigate the possible temporal cues available, the responses of fibers in the cochlear nerve of the anesthetized guinea pig to synthetic vowels were recorded at a range of sound levels up to 95 dB SPL. The stimuli were the single vowels /i/ [fundamental frequency (f0) 125 Hz], /a/ (f0, 100 Hz), and /c/ (f0, 100 Hz) and the double vowels were /a(100),i(125)/ and /c(100),i(125)/. Histograms synchronized to the period of the double vowels were constructed, and locking of the discharge to individual harmonics was estimated from them by Fourier transformation. One possible cue for identifying the f0's of the constituents of a double vowel is modulation of the neural discharge with a period of 1/f0. Such modulation was found at frequencies between the formant peaks of the double vowel, with modulation at the periods of 100 and 125 Hz occurring at different places in the fiber array. Generation of a population response based on synchronized responses [average localized synchronized rate (ALSR): see Young and Sachs [J. Acoust. Soc. Am. 66, 1381-1403 (1979)] allowed estimation of the f0's by a variety of methods and subsampling the population response at the harmonics of the f0 of the constituent vowel achieved a good reconstruction of its spectrum. Other analyses using interval histograms and autocorrelation, which overcome some problems associated with the ALSR approach, also allowed f0 identification and vowel segregation. The present study has demonstrated unequivocally that the timing of the impulses in auditory-nerve fibers provides copious possible cues for the identification of the fundamental frequencies and spectra associated with each of the constituents of double vowels.

The effect of vibrato on the recognition of masked vowels

Five experiments on the identifiability of synthetic vowels masked by wideband sounds are reported. In each experiment, identification thresholds (signal/masker ratios, in decibels) were measured for two versions of four vowels: a vibrated version, in which FO varied sinusoidally around 100 Hz; and a steady version, in which F0 was fixed at 100 Hz. The first three experiments were performed on naive subjects. Experiment 1 showed that for maskers consisting of bursts of pink noise, vibrato had no effect on thresholds. In Experiment 2, where the maskers were periodic pulse trains with an F0 randomly varied between 120 and 140 Hz from trial to trial, vibrato slightly improved thresholds when the sound pressure level of the maskers was 40 dB, but had no effect for 65-dB maskers. In Experiment 3, vibrated rather than steady pulse trains were used as maskers; when these maskers were at 40 dB, the vibrated versions of the vowels were slightly less identifiable than their steady versions; but, as in Experiment 2, vibrato had no effect when the maskers were at 65 dB. Experiment 4 showed that the unmasking effect of vibrato found in Experiment 2 disappeared in subjects trained in the identification task. Finally, Experiment 5 indicated that in trained listeners, vibrato had no influence on identification performance even when the maskers and the vowels had synchronous onsets and offsets. We conclude that vibrating a vowel masked by a wideband sound can affect its identification threshold, but only for tonal maskers and in untrained listeners. This effect of vibrato should probably be considered as a Gestalt phenomenon originating from central auditory mechanisms.

Modeling the perception of concurrent vowels: Vowels with different fundamental frequencies

If two vowels with different fundamental frequencies (fo's) are presented simultaneously and monaurally, listeners often hear two talkers producing different vowels on different pitches. This paper describes the evaluation of four computational models of the auditory and perceptual processes which may underlie this ability. Each model involves four stages: (i) frequency analysis using an "auditory" filter bank, (ii) determination of the pitches present in the stimulus, (iii) segregation of the competing speech sources by grouping energy associated with each pitch to create two derived spectral patterns, and (iv) classification of the derived spectral patterns to predict the probabilities of listeners' vowel-identification responses. The "place" models carry out the operations of pitch determination and spectral segregation by analyzing the distribution of rms levels across the channels of the filter bank. The "place-time" models carry out these operations by analyzing the periodicities in the waveforms in each channel. In their "linear" versions, the place and place-time models operate directly on the waveforms emerging from the filters. In their "nonlinear" versions, analogous operations are applied to the output of an additional stage which applied a compressive nonlinearity to the filtered waveforms. Compared to the other three models, the nonlinear place-time model provides the most accurate estimates of the fo's of paris of concurrent synthetic vowels and comes closest to predicting the identification responses of listeners to such stimuli. Although the model has several limitations, the results are compatible with the idea that a place-time analysis is used to segregate competing sound sources.

Vowel nasalization: An acoustic and perceptual study of natural and synthetic vowels

The purpose of the present study was to determine the acoustic cues used by speakers to nasalize vowels. These cues were then used to synthesize nasal vowels. The synthetic vowels, along with their naturally produced counterparts, were presented to a group of judges to rate vowel identity, naturalness, and the presence or absence of nasality. Speech samples were obtained from ten speakers who produced the vowels /i, u, æ, a, ɔ, ʌ/in a variety of non‐nasal and nasal contexts. DFT analyses were performed on the utterances. Differences between the non‐nasal and nasal vowels were studied by comparing pole frequency, amplitude and band‐width values, zero frequency values, F0 values, first‐harmonic amplitude in relation to first‐formant amplitude (H1‐A1) values, spectral tilt, and open‐quotient values. The 1988 version of the Klatt synthesizer was used to synthesize nasality in vowels based upon the cues identified. Accuracy of the judges perceptions of the [+ nasal] feature in the synthesized vowels was dependent on the presence of a nasal pole‐zero pair in the vowel spectrum. Frequency of the nasal pole‐zero pair varied with the specific vowel being synthesized.

Difference limens for synthetic vowel spectra

Difference limens (DLs) along synthetic vowel continua were measured as distances in the auditory-perceptual space proposed by J. D. Miller [J. Acoust. Soc. Am. 85, 2114–2134 (1989)]. The continua represent straight lines that lie parallel with one of the axes of the space (x′, y′, and z′). Groups of six continua share a common center value, or reference point. Movement along these continua result in distinct patterns of frequency change for F1, F2, and F3 relative to the reference point formant values. These patterns vary with the direction and axis of movement. Reference points were selected from the interiors of ten vowel zones and seven boundary areas between vowel zones. An adaptive up—down procedure employing a cued, two-alternative, forced-choice (2AFC) task was utilized to estimate the 79.5% correct point along each continuum. Vowel tokens representing the reference points served as the cue stimuli. DLs were estimated twice for each continuum from each of four subjects. In general, the results of this experiment reflect smaller DLs for vowel formant frequencies than have been found in the past and suggest that while an overall average DL for distance in the auditory-perceptual space may be estimated, DLs vary significantly with the axis of movement. Additionally, no significant difference was found between continua associated with vowel centers and vowel boundaries. These results will be compared with similar results estimating DLs for single-formant variation in vowels. [Work supported by NIDCD.]

Temporal Effects in Simultaneous Masking by Vowel and Consonant-Vowel Maskers

Temporal effects in simultaneous masking were studied using synthetic vowel (V) and consonant-vowel (CV) maskers. For the steady-state V maskers (/i,a,u/), signals were presented at the beginning or in the temporal center of the masker. The masking patterns generally reflected the formant differences among the vowels, and the formant structure of each V masker was more clearly revealed when the signal was presented in the temporal center of the masker. For the CV maskers (/bi,gi/), signals were presented at the beginning of the (consonant portion of the) masker, at the beginning of the vowel portion of the masker, or in the temporal center of the masker. The second-formant difference between the maskers (observed acoustically at their onset) was generally revealed in the masking patterns when the signal was presented at the beginning of the consonant; this difference in the masking patterns was also present, to a lesser extent, when the signal was presented at the beginning of the vowel, where the two maskers were identical acoustically. The masking patterns for the two CV maskers were virtually identical when the signal was presented in the temporal center of the masker. These data extend previous tone-on-tone masking data and suggest that the auditory system requires a certain amount of time to represent most accurately the acoustic spectrum of both steady-state and dynamic complex maskers.

Speech audiometric trial using synthetic vowels produced with DECtalk

A speech audiometric trial is described for testing the degree of hearing loss and its pathogenesis. Test vowels and distorted speech sounds were synthesized with DECtalk for use in speech audiometric tests. Results showed that 100% articulation can be obtained with undistorted synthesized vowels, while reasonable articulation scores can be obtained with distorted synthetic speech signals. This method could solve some of the problems inherent in use of recorded natural Japanese speech sounds as acoustic stimuli during otological examinations.

Differential sensitivity of newborns to synthetic vowel stimuli

Three-day-old babies were presented with synthetic speech stimuli and some no-sound trials. Three factors of the speech stimuli were examined: vowel quality change, pitch (F0) change, and stimulus presentation rate. The motor behaviour of the neonates was assessed using a multi-response observation method. The babies responded more when the vowel quality changed than when it did not. In some conditions, pitch variation and stimulus presentation rate also affected d'. The work extends previous findings on newborn auditory sensitivity showing differential responding to steady state vowels and that alterations in F0 can affect behaviour. The method has advantages in permitting several stimuli to be compared as well as studying neonates in states of arousal typical of perinatal experience.

A pitch perception study for a cochlear implant user with residual hearing in the contralateral ear

In order to combine electrical and acoustic signals for multichannel cochlear implant users with residual hearing in the unimplanted ear, it is important to determine the degree of overlap in pitch sensations on the two sides. A pitch scaling experiment was performed maintaining constant and identical repetition rates on both the implanted (left) and normal (right) ear. The acoustic signals consisted of bandpass‐filtered pulses {with center frequencies ranging from 350–850 Hz). These were transmitted in synchrony with electrical pulses to one of five selected bipolar electrode pairs spanning 15.75 mm of the scala tympani (from approximately 2.25 to 18 mm beyond the base). The spacing between the two electrodes of each bipolar pair was 1.5 min. The experiment was repeated for repetition rates of 112.5 and 205 Hz. Pitch sensations on the left (electrical) side were reported to be as low as, and at times even lower than, on the right (acoustic) side. All signals were sealed with respect to an acoustic reference stimulus that had a bandpass filter center frequency at 450 Hz. Synthetic vowels were derived from the pitch scaling results and acoustic strategies only as well as combined electrical and acoustic strategies were tested. It was found that the best results were obtained from a combined electrical and acoustic strategy where the electrode information is chosen such that pitch perception overlap with the acoustic side is minimized. [Work supported by NIH Grant NS21027.]

Perception of synthetic vowels: A comparison of several classification schemes

Synthetic vowel tokens (1725) were randomly presented twice to eight naive speakers of Midwestern American English for classification as one of 12 vowel categories, /IY,IH,EH,EY,AE,AA,AH,AO,OW,UH, UW,ER/. Subjects rated the certainty of their responses on a scale from one (very unsure) to five (very sure). The vowels were synthesized in null context and utilized a male F0 contour. The frequency values of F1, F2, and F3 were selected such that the entire area where vowels may be represented in the auditory‐perceputal space [J. D. Miller, J. Acoust. Soc. Am. 85, 2114–2134 (1989)] was equidistantly sampled. The results of this experiment support the view that F1 and F2 are the primary determinants in the perception of nonretroflex vowels. In addition, monophthongal versions of the diphthongs /EY/ and /OW/ may be distinctly classified. The results will be presented graphically as target zones constructed on the basis of the plurality identifications for each token. These target zones are abutting and nonoverlapping, and correctly classify 99.9% of the plurality judgments (less 48 tokens where ties occurred) and 75% of the 27 600 total judgments. Other vowel classification schemes will also be compared for their accuracy in classifying the results of this experiment, as well as results from other studies. [Work supported by NINCD.]