This study constitutes a large-scale comparative analysis of acoustic cues for classification of place of articulation in fricatives. To date, no single metric has been found to classify fricative place of articulation with a high degree of accuracy. This study presents spectral, amplitudinal, and temporal measurements that involve both static properties (spectral peak location, spectral moments, noise duration, normalized amplitude, and onset frequency) and dynamic properties (relative amplitude and locus equations). While all cues (except locus equations) consistently serve to distinguish sibilant from nonsibilant fricatives, the present results indicate that spectral peak location, spectral moments, and both normalized and relative amplitude serve to distinguish all four places of fricative articulation. These findings suggest that these static and dynamic acoustic properties can provide robust and unique information about all four places of articulation, despite variation in speaker, vowel context, and voicing.
1 December 1999 in The Journal of the Acoustical Society of America
Auditory training has been shown to be effective in the identification of non-native segmental distinctions. In this study, it was investigated whether such training is applicable to the acquisition of non-native suprasegmental contrasts, i.e., Mandarin tones. Using the high-variability paradigm, eight American learners of Mandarin were trained in eight sessions during the course of two weeks to identify the four tones in natural words produced by native Mandarin talkers. The trainees’ identification accuracy revealed an average 21% increase from the pretest to the post-test, and the improvement gained in training was generalized to new stimuli (18% increase) and to new talkers and stimuli (25% increase). Moreover, the six-month retention test showed that the improvement was retained long after training by an average 21% increase from the pretest. The results are discussed in terms of non-native suprasegmental perceptual modification, and the analogies between L2 acquisition processes at the segmental and suprasegmental levels.
28 January 2003 in The Journal of the Acoustical Society of America
Training American listeners to perceive Mandarin tones has been shown to be effective, with trainees’ identification improving by 21%. Improvement also generalized to new stimuli and new talkers, and was retained when tested six months after training [Y. Wang et al., J. Acoust. Soc. Am. 106, 3649–3658 (1999)]. The present study investigates whether the tone contrasts gained perceptually transferred to production. Before their perception pretest and after their post-test, the trainees were recorded producing a list of Mandarin words. Their productions were first judged by native Mandarin listeners in an identification task. Identification of trainees’ post-test tone productions improved by 18% relative to their pretest productions, indicating significant tone production improvement after perceptual training. Acoustic analyses of the pre- and post-training productions further reveal the nature of the improvement, showing that post-training tone contours approximate native norms to a greater degree than pretraining tone contours. Furthermore, pitch height and pitch contour are not mastered in parallel, with the former being more resistant to improvement than the latter. These results are discussed in terms of the relationship between non-native tone perception and production as well as learning at the suprasegmental level.
Most theories of categorization emphasize how continuous perceptual information is mapped to categories. However, equally important are the informational assumptions of a model, the type of information subserving this mapping. This is crucial in speech perception where the signal is variable and context dependent. This study assessed the informational assumptions of several models of speech categorization, in particular, the number of cues that are the basis of categorization and whether these cues represent the input veridically or have undergone compensation. We collected a corpus of 2,880 fricative productions (Jongman, Wayland, & Wong, 2000) spanning many talker and vowel contexts and measured 24 cues for each. A subset was also presented to listeners in an 8AFC phoneme categorization task. We then trained a common classification model based on logistic regression to categorize the fricative from the cue values and manipulated the information in the training set to contrast (a) models based on a small number of invariant cues, (b) models using all cues without compensation, and (c) models in which cues underwent compensation for contextual factors. Compensation was modeled by computing cues relative to expectations (C-CuRE), a new approach to compensation that preserves fine-grained detail in the signal. Only the compensation model achieved a similar accuracy to listeners and showed the same effects of context. Thus, even simple categorization metrics can overcome the variability in speech when sufficient information is available and compensation schemes like C-CuRE are employed.
1 September 1997 in The Journal of the Acoustical Society of America
This study investigated speaker normalization in perception of Mandarin tone 2 (midrising) and tone 3 (low-falling–rising) by examining listeners’ use of F0 range as a cue to speaker identity. Two speakers were selected such that tone 2 of the low-pitched speaker and tone 3 of the high-pitched speaker occurred at equivalent F0 heights. Production and perception experiments determined that turning point (or inflection point of the tone), and ΔF0 (the difference in F0 between onset and turning point) distinguished the two tones. Three tone continua varying in either turning point, ΔF0, or both acoustic dimensions, were then appended to a natural precursor phrase from each of the two speakers. Results showed identification shifts such that identical stimuli were identified as low tones for the high precursor condition, but as high tones for the low precursor condition. Stimuli varying in turning point showed no significant shift, suggesting that listeners normalize only when the precursor varies in the same dimension as the stimuli. The magnitude of the shift was greater for stimuli varying only in ΔF0, as compared to stimuli varying in both turning point and ΔF0, indicating that normalization effects are reduced for stimuli more closely matching natural speech.
1 June 2009 in The Journal of the Acoustical Society of America
Speakers can adopt a speaking style that allows them to be understood more easily in difficult communication situations, but few studies have examined the acoustic properties of clearly produced consonants in detail. This study attempts to characterize the adaptations in the clear production of American English fricatives in a carefully controlled range of communication situations. Ten female and ten male talkers produced fricatives in vowel-fricative-vowel contexts in both a conversational and a clear style that was elicited by means of simulated recognition errors in feedback received from an interactive computer program. Acoustic measurements were taken for spectral, amplitudinal, and temporal properties known to influence fricative recognition. Results illustrate that (1) there were consistent overall style effects, several of which (consonant duration, spectral peak frequency, and spectral moments) were consistent with previous findings and a few (notably consonant-to-vowel intensity ratio) of which were not; (2) specific acoustic modifications in clear productions of fricatives were influenced by the nature of the recognition errors that prompted the productions and were consistent with efforts to emphasize potentially misperceived contrasts both within the English fricative inventory and based on feedback from the simulated listener; and (3) talkers differed widely in the types and magnitude of all modifications.