Synthetic Vowels Research Articles

F1/F2의 변화가 한국어 /오/, /우/ 모음의 지각판별에 미치는 영향

This study examined the perception of two Korean vowels using F1/F2 manipulated synthetic vowels. Previous studies indicated that there is an overlap between the acoustic spaces of Korean /o/ and /u/ in terms of the first two formants. A continuum of eleven synthetic vowels were used as stimuli. The experiment consisted of three tasks: an /o/ identification task (Yes-no), an /u/ identification task (Yes-no), and a forced choice identification task (/o/-/u/). ROC(Receiver Operating Characteristic) analysis and logistic regression were performed to calculate the boundary criterion of the two vowels along the stimulus continuum, and to predict the perceptual judgment on F1 and F2. The result indicated that the location between stimulus no.5 (F1 = 342Hz, F2 = 691Hz) and no.6 (F1 = 336Hz, F2 = 700Hz) was estimated as a perceptual boundary region between /o/ and /u/, while stimulus no.0 (F1=405Hz, F2=666Hz) and no.10 (F1=321Hz, F2=743Hz) were at opposite ends of the continuum. The influence of F2 was predominant over F1 on the perception of the vowel categories.

Formant frequency estimation of high-pitched vowels using weighted linear prediction

All-pole modeling is a widely used formant estimation method, but its performance is known to deteriorate for high-pitched voices. In order to address this problem, several all-pole modeling methods robust to fundamental frequency have been proposed. This study compares five such previously known methods and introduces a technique, Weighted Linear Prediction with Attenuated Main Excitation (WLP-AME). WLP-AME utilizes temporally weighted linear prediction (LP) in which the square of the prediction error is multiplied by a given parametric weighting function. The weighting downgrades the contribution of the main excitation of the vocal tract in optimizing the filter coefficients. Consequently, the resulting all-pole model is affected more by the characteristics of the vocal tract leading to less biased formant estimates. By using synthetic vowels created with a physical modeling approach, the results showed that WLP-AME yields improved formant frequencies for high-pitched sounds in comparison to the previously known methods (e.g., relative error in the first formant of the vowel [a] decreased from 11% to 3% when conventional LP was replaced with WLP-AME). Experiments conducted on natural vowels indicate that the formants detected by WLP-AME changed in a more regular manner between repetitions of different pitch than those computed by conventional LP.

Acoustic coupling during incomplete glottal closure and its effect on the inverse filtering of oral airflow

Inverse filtering of oral airflow using closed-phase linear prediction is expected to preserve the effects of source-filter interactions in the glottal airflow pulse. Under incomplete glottal closure, the glottal airflow estimation is more challenging due to a lowered glottal impedance, increased subglottal coupling, and violated all-pole assumption. To account for these effects, a model-based inverse filtering scheme allowing for coupling between glottis and upper and lower airways was developed. Acoustic transmission in the tracts used a frequency-domain transmission line. A linearized, time-varying expression was used for the glottal impedance, along with a dipole representation. Synthetic vowels sounds and actual recordings were used to evaluate the proposed scheme. Subject-specific model parameters were obtained from simultaneous aerodynamic, acoustic, and high-speed videoendoscopic recordings of normal subjects uttering vowels with various degrees of glottal closure. Results illustrated that, even under incomplete glottal closure, the airflow entering the vocal tract preserved source-filter interactions and was comparable to that obtained using closed-phase linear prediction. The scheme also yielded an uncoupled glottal airflow that exhibited a clear pulse de-skewing, making it proportional to the glottal area. Cases with larger glottal gaps exhibited lower mean impedances and less pulse skewing, with airflow estimates proportional to the transglottal pressure drop.

Rate and onset cues can improve cochlear implant synthetic vowel recognition in noise

Understanding speech-in-noise is difficult for most cochlear implant (CI) users. Speech-in-noise segregation cues are well understood for acoustic hearing but not for electric hearing. This study investigated the effects of stimulation rate and onset delay on synthetic vowel-in-noise recognition in CI subjects. In experiment I, synthetic vowels were presented at 50, 145, or 795 pulse/s and noise at the same three rates, yielding nine combinations. Recognition improved significantly if the noise had a lower rate than the vowel, suggesting that listeners can use temporal gaps in the noise to detect a synthetic vowel. This hypothesis is supported by accurate prediction of synthetic vowel recognition using a temporal integration window model. Using lower rates a similar trend was observed in normal hearing subjects. Experiment II found that for CI subjects, a vowel onset delay improved performance if the noise had a lower or higher rate than the synthetic vowel. These results show that differing rates or onset times can improve synthetic vowel-in-noise recognition, indicating a need to develop speech processing strategies that encode or emphasize these cues.

Brainstem Auditory Responses to Resolved and Unresolved Harmonics of a Synthetic Vowel in Quiet and Noise

This study investigated speech auditory brainstem responses (speech ABR) with variants of a synthetic vowel in quiet and in background noise. Its objectives were to study the noise robustness of the brainstem response at the fundamental frequency F0 and at the first formant F1, evaluate how the resolved/unresolved harmonics regions in speech contribute to the response at F0, and investigate the origin of the response at F0 to resolved and unresolved harmonics in speech. In total, 18 normal-hearing subjects (11 women, aged 18-33 years) participated in this study. Speech ABRs were recorded using variants of a 300 msec formant-synthesized /a/ vowel in quiet and in white noise. The first experiment employed three variants containing the first three formants F1 to F3, F1 only, and F2 and F3 only with relative formant levels following those reported in the literature. The second experiment employed three variants containing F1 only, F2 only, and F3 only, with the formants equalized to the same level and the signal-to-noise ratio (SNR) maintained at -5 dB. Overall response latency was estimated, and the amplitude and local SNR of the envelope following response at F0 and of the frequency following response at F1 were compared for the different stimulus variants in quiet and in noise. The response at F0 was more robust to noise than that at F1. There were no statistically significant differences in the response at F0 caused by the three stimulus variants in both experiments in quiet. However, the response at F0 with the variant dominated by resolved harmonics was more robust to noise than the response at F0 with the stimulus variants dominated by unresolved harmonics. The latencies of the responses in all cases were very similar in quiet, but the responses at F0 due to resolved and unresolved harmonics combined nonlinearly when both were present in the stimulus. Speech ABR has been suggested as a marker of central auditory processing. The results of this study support earlier work on the differential susceptibility to noise of the F0 and F1 components of the evoked response. In the case of F0, the results support the view that in speech, the pitch of resolved harmonics and that of unresolved harmonics are processed in different but interacting pathways that converge in the upper brainstem. Pitch plays an important role in speech perception, and speech ABR can offer a window into the neural extraction of the pitch of speech and how it may change with hearing impairment.

Phonetically Trained and Untrained Adults' Transcription of Place of Articulation for Intervocalic Lingual Stops With Intermediate Acoustic Cues

In this study, the authors aimed to investigate how listener training and the presence of intermediate acoustic cues influence transcription variability for conflicting cue speech stimuli. Twenty listeners with training in transcribing disordered speech, and 26 untrained listeners, were asked to make forced-choice labeling decisions for synthetic vowel-consonant-vowel (VCV) sequences "a doe" (/do/) and "a go" (/go/). Both the VC and CV transitions in these stimuli ranged through intermediate positions, from appropriate for /d/ to appropriate for /g/. Both trained and untrained listeners gave more weight to the CV transitions than to the VC transitions. However, listener behavior was not uniform: The results showed a high level of inter- and intratranscriber inconsistency, with untrained listeners showing a nonsignificant tendency to be more influenced than trained listeners by CV transitions. Listeners do not assign consistent categorical labels to the type of intermediate, conflicting transitional cues that were present in the stimuli used in the current study and that are also present in disordered articulations. Although listener inconsistency in assigning labels to intermediate productions is not increased as a result of phonetic training, neither is it reduced by such training.

Objective measurement of physiological signal-to-noise gain in the brainstem response to a synthetic vowel

ObjectiveThis work investigates auditory speech processing in normal listeners through measurement of brainstem responses to the synthetic vowel /a/. MethodsThe vowel is presented in quiet and in continuous white noise with different signal-to-noise ratios (SNR) of +5, 0, −5, and −10dB. ResultsIn the presence of noise, transient response waves V and A are delayed when compared to those evoked in quiet, whereas the amplitude of wave V and the steepness of the slope between waves V and A are strongly reduced. The spectral component of the steady-state evoked response corresponding to the fundamental frequency (F0) of the vowel shows significantly greater amplitude and local SNR in the less severe noise conditions compared to the quiet condition. Such increases of the amplitude and SNR were not observed for the spectral component corresponding to the first formant of the vowel (F1). ConclusionsResults suggest that, at F0, both local noise suppression and signal enhancement contribute to the SNR gain. There is suppression of local noise near F1, but no signal enhancement. SignificanceThe physiological SNR gain was estimated to be approximately +12dB at both F0 and F1, as stimulus SNR was reduced from +5 to −10dB.

A cross-linguistic study of the effect of perceived gender on the categorization of children's vowels

Previous research has shown that listeners classify acoustically gender-ambiguous vowels differently depending on whether they believe the speaker to be a man or a woman (Johnson, Strand, & D'Imperio, 1999). In real-world listening situations, this tendency could be especially pronounced in the perception of vowels produced by children, as children's voices are inherently gender-ambiguous (Perry, Ohde, & Ashmead, 2001). Given Johnson et al.'s findings, we would predict that adults would categorize children's vowels differently depending on whether they believe the speaker to bes male or female. To examine this hypothesis, a set of experiments was conducted in which native speakers of Cantonese, Japanese, and English categorized a series of synthetic vowels (Menard et al., 2006). These were generated using an articulatory synthesizer, and were meant to represent the vocal tracts of newborn children, 2-, 4-, 5-, 10-, 16-, and 21-year olds. Adults categorized these vowels and provided judgments of the perceived gender and age of the speakers. Preliminary results suggest that speakers of English categorize vowels differently depending on whether they judge the child to be male or female. This tendency is especially marked for the most gender-ambiguous stimulus set, those based on the 10-year-old vocal tract.

Forward Masking of Dynamic Acoustic Intensity: Effects of Intensity Region and End-Level

Overestimation of loudness change typically occurs in response to up-ramp auditory stimuli (increasing intensity) relative to down-ramps (decreasing intensity) matched on frequency, duration, and end-level. In the experiment reported, forward masking is used to investigate a sensory component of up-ramp overestimation: persistence of excitation after stimulus presentation. White-noise and synthetic vowel 3.6 s up-ramp and down-ramp maskers were presented over two regions of intensity change (40-60 dB SPL, 60-80 dB SPL). Three participants detected 10 ms 1.5 kHz pure tone signals presented at masker-offset to signal-offset delays of 10, 20, 30, 50, 90, 170 ms. Masking magnitude was significantly greater in response to up-ramps compared with down-ramps for masker-signal delays up to and including 50 ms. When controlling for an end-level recency bias (40-60 dB SPL up-ramp vs 80-60 dB SPL down-ramp), the difference in masking magnitude between up-ramps and down-ramps was not significant at each masker-signal delay. Greater sensory persistence in response to up-ramps is argued to have minimal effect on perceptual overestimation of loudness change when response biases are controlled. An explanation based on sensory adaptation is discussed.

The Continuity Illusion Does Not Depend on Attentional State: fMRI Evidence from Illusory Vowels

We investigate whether the neural correlates of the continuity illusion, as measured using fMRI, are modulated by attention. As we have shown previously, when two formants of a synthetic vowel are presented in an alternating pattern, the vowel can be identified if the gaps in each formant are filled with bursts of plausible masking noise, causing the illusory percept of a continuous vowel ("Illusion" condition). When the formant-to-noise ratio is increased so that noise no longer plausibly masks the formants, the formants are heard as interrupted ("Illusion Break" condition) and vowels are not identifiable. A region of the left middle temporal gyrus (MTG) is sensitive both to intact synthetic vowels (two formants present simultaneously) and to Illusion stimuli, compared to Illusion Break stimuli. Here, we compared these conditions in the presence and absence of attention. We examined fMRI signal for different sound types under three attentional conditions: full attention to the vowels; attention to a visual distracter; or attention to an auditory distracter. Crucially, although a robust main effect of attentional state was observed in many regions, the effect of attention did not differ systematically for the illusory vowels compared to either intact vowels or to the Illusion Break stimuli in the left STG/MTG vowel-sensitive region. This result suggests that illusory continuity of vowels is an obligatory perceptual process, and operates independently of attentional state. An additional finding was that the sensitivity of primary auditory cortex to the number of sound onsets in the stimulus was modulated by attention.

Formant frequencies, vowel identity, and the perceived relative tallness of synthetic speakers

Listeners can make consistent judgments regarding the tallness of speakers [Rendall etal., J. Exp. Psych: Human Percep. Perform. 33, 1208 (2007)]. These judgments are informed by the f0 and formant frequencies (FFs) of a speaker's voice. However, FFs are also cues to vowel identity, such that a small speaker producing an /u/ might have lower average FFs than a larger speaker producing an /æ/. Do listeners use absolute FFs to judge speaker tallness, or do they “correct” for phonetic identity and consider the FFs of a vowel relative to those expected for that category? To test this, a series of synthetic vowels (/i æ u/) with different FF scalings and different numbers of formants (2–5) were created. These scalings were intended to replicate speakers of different vocal tract lengths (i.e., sizes). Participants were presented with pairs of vowels and asked to indicate which vowel sounded like it had been produced by a taller speaker. Results indicate that listeners consider both absolute and phonetically “corrected” FF information and that formants higher than F3 greatly reduce listener's reliance on absolute F1 and F2 information in making speaker size judgments.

Contribution of Resolved and Unresolved Harmonic Regions to Brainstem Speech-Evoked Responses in Quiet and in Background Noise.

Speech auditory brainstem responses (speech ABR) reflect activity that is phase-locked to the harmonics of the fundamental frequency (F0) up to at least the first formant (F1). Recent evidence suggests that responses at F0 in the presence of noise are more robust than responses at F1, and are also dissociated in some learning-impaired children. Peripheral auditory processing can be broadly divided into resolved and unresolved harmonic regions. This study investigates the contribution of these two regions to the speech ABR, and their susceptibility to noise. We recorded, in quiet and in background white noise, evoked responses in twelve normal hearing adults in response to three variants of a synthetic vowel: i) Allformants, which contains all first three formants, ii) F1Only, which is dominated by resolved harmonics, and iii) F2&F3Only, which is dominated by unresolved harmonics. There were no statistically significant differences in the response at F0 due to the three variants of the stimulus in quiet, nor did the noise affect this response with the Allformants and F1Only variants. On the other hand, the response at F0 with the F2&F3Only variant was significantly weaker in noise than with the two other variants (p<0.001). With the response at F1, there was no difference with the Allformants and F1Only variants in quiet, but was expectedly weaker with the F2&F3Only variant (p<0.01). The addition of noise significantly weakened the response at F1 with the F1Only variant (p<0.05), but this weakening only tended towards significance with the Allformants variant (p=0.07). The results of this study indicate that resolved and unresolved harmonics are processed in different but interacting pathways that converge in the upper brainstem. The results also support earlier work on the differential susceptibility of responses at F0 and F1 to added noise.

A model for the prediction of breathiness in vowels

The perception of breathiness in vowels is cued by multiple acoustic cues, including changes in aspiration noise (AH) and the open quotient (OQ) [Klatt and Klatt, J. Acoust. Soc. Am. 87(2), 820-857 (1990)]. A loudness model can be used to determine the extent to which AH masks the harmonic components in voice. The resulting "partial loudness" (PL) and loudness of AH ["noise loudness" (NL)] have been shown to be good predictors of perceived breathiness [Shrivastav and Sapienza, J. Acoust. Soc. Am. 114(1), 2217-2224 (2003)]. The levels of AH and OQ were systematically manipulated for ten synthetic vowels. Perceptual judgments of breathiness were obtained and regression functions to predict breathiness from the ratio of NL to PL (η) were derived. Results show that breathiness can be modeled as a power function of η. The power parameter of this function appears to be affected by the fundamental frequency of the vowel. A second experiment was conducted to determine if the resulting power function could estimate breathiness in a different set of voices. The breathiness of these stimuli, both natural and synthetic, was determined in a listening test. The model estimates of breathiness were highly correlated with perceptual data but the absolute predicted values showed some discrepancies.

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.

Learning the Dynamical System Behind Sensory Data

This letter looks at the physics behind sensory data by identifying the parameters that govern the physical system and estimating them from sensory observations. We extend Takens's delay-embedding theorem to the dynamical systems controlled by parameters. An embedding of the product space of the phase and the parameter spaces of the dynamical system can be obtained by the delay-embedding map, provided that the parameter of the dynamical system changes slowly. The reconstruction error is bounded for slowly varying parameters. A manifold learning technique is applied to the embedding obtained to extract a low-dimensional global coordinate system representing the product space. The phase space of the deterministic dynamics can be contracted by using the adjacency relationship in time, which enables recovery of only the parameter space. As examples, the manifolds of synthetic and real-world vowels with time-varying fundamental frequency (F(0)) are analyzed, and the F(0) contours are extracted by an unsupervised algorithm. Experimental results show that the proposed method leads to robust performance under various noise conditions and rapid changes of F(0) compared with the current state-of-the-art F(0) estimation algorithms.

Accuracy of formant measurement for synthesized vowels using the reassigned spectrogram and comparison with linear prediction

This brief report describes a small study which was undertaken with nine synthetic vowel tokens, in an effort to demonstrate the validity of the reassigned spectrogram as a formant measurement tool. The reassigned spectrogram's performance is also compared with that of a typical pitch-asynchronous linear predictive analysis and is found to be superior. In this study, reassigned spectrograms were further processed to highlight the formants and then were used to measure these synthetic vowel formants generally to within 0.5% of their known true values, far surpassing the accuracy of a typical linear predictive analysis procedure which was inaccurate by as much as 17%. The overall accuracy of reassigned spectrographic formant measurement is thus demonstrated in these cases.

Identification of synthetic vowels based on a time-varying model of the vocal tract area function

The purpose of this study was to conduct an identification experiment with synthetic vowels based on the same sets of speaker-dependent area functions as in Bunton and Story [(2009) J. Acoust. Soc. Am. 125, 19-22], but with additional time-varying characteristics that are more representative of natural speech. The results indicated that vowels synthesized using an area function model that allows for time variation of the vocal tract shape and includes natural vowel durations were more accurately identified for 7 of 11 English vowels than those based on static area functions.

Neural timing nets for fundamental frequency (F0)-based auditory scene analysis.

Neural timing nets are idealized networks of delay lines, coincidence detectors, and adaptive processing elements that operate in the time domain on temporally coded signals to compare, extract, and separate auditory objects [Cariani, Neural Networks 14, 737–753 (2001); J. New Music Res. 30, 107–135; IEEE Trans. Neural Netw. 15(5) (2004)]. Timing nets constitute an alternative, potential mode of neural signal processing in which information resides in neural signals rather than in patterns of activated elements. Recurrent timing nets with delay loops act as dense arrays of recursive, comblike filters to effect a period-by-period analysis that builds up and separates component invariant time patterns with different fundamentals (F0s). Using both linear and nonlinear processing rules, the latter were used to process and separate synthetic double vowels, running speech, and polyphonic musical excerpts, with varying results. Relations to, and combinations with, processing strategies based on autocorrelation and all-order interspike intervals, adaptive comb filtering, correlogram duplex analysis, cancellation, Fourier zero-crossings, and demodulation, with and without prior bandpass filtering, are discussed. Bottom-up/top-down mechanisms for dynamic facilitation of lower level temporal processing loops are also considered. [Work supported by AFOSR FA9550-09-1-0119.]

The relationship between fundamental frequency and vowel quality.

There is disagreement over the role fundamental frequency (f0) plays in the determination of vowel quality. Some claim [e.g., D. Smith et al., J. Acoust. Soc. Am. 117, 305–318 (2005)] that changes in f0 do not affect vowel quality at all. Others claim there is relationship between f0 and vowel quality, whether direct or indirect [see T. M. Nearey and P. F. Assmann, in Experimental Approaches to Phonology, edited by M. J. Solé and P. S. Beddor (Oxford University Press, Oxford, 2007), pp. 246–269 for a review]. To test these theories, a series of experiments was carried out in which participants were asked to make simultaneous speaker and vowel quality judgments. Participants were presented with a synthetic vowel continuum spanning from [æ] to [ʌ] matched with several different f0s and higher formants. Participants were asked to rate vowel quality on a continuous scale ranging from completely [æ] to completely [ʌ] and to indicate the size and gender of the speaker. Results indicate that there is a complicated, indirect relationship between f0 and vowel quality, and that shifts in vowel quality caused by changes in f0 are a result of changes in assumed speaker properties.

Analysis of categorical response data: Use logistic regression rather than endpoint-difference scores or discriminant analysis

Example of a typical second-language (L2) speech perception experiment: Synthetic vowel stimuli from a two-dimensional grid of points in which acoustic properties vary systematically in duration and spectral properties are classified as English /i/ or /I/ by L2-English listeners. In a number of studies, the data from such experiments have been analyzed using endpoint-difference scores or discriminant analysis. The current letter describes theoretical problems inherent in the first procedure in general, and in the application of the second procedure to data of this type in particular. Logistic regression is proposed as an alternative, which does not suffer from these problems.