Shuangyu Chang

Syllable-proximity evaluation in automatic speech recognition using fuzzy measures and a fuzzy integral

This paper describes a syllable-proximity evaluation task within the context of an automatic speech recognition application. This task is well suited to a multiple-information aggregation framework in which preliminary evaluations of separate information sources are combined to produce a more accurate and reliable overall evaluation than would otherwise be the case. An aggregation operator using fuzzy measures and a fuzzy integral is adopted that possesses a number of desirable properties and the fuzzy-measure parameters can be automatically learned from training data by re-casting the syllable-proximity evaluation as a classification problem. Experiments performed on spontaneous speech material demonstrate that the fuzzy-integration-based aggregation approach has many advantages over alternative techniques in terms of both performance and interpretability of the system.

Application of Fuzzy-Integration-Based Multiple-Information Aggregation in Automatic Speech Recognition

Many real-world problems can be cast into a multiple-information aggregation framework where preliminary evaluations of separate information sources are combined to produce more accurate and reliable evaluation than would otherwise be the case. In this paper we describe a syllable-proximity evaluation problem in automatic speech recognition that fits well into this aggregation framework. A fuzzy-integration-based approach is adopted as the aggregation operator and a gradient-based algorithm is described for learning parameters automatically from training data. Experiments using spontaneous speech material demonstrate that the fuzzy-integration-based aggregation approach has many advantages over other techniques in terms of both performance and interpretability of the system.

Beyond the phoneme: a juncture-accent model of spoken language

Phonemic models of spoken language are incapable of accommodating the patterns of pronunciation variation observed in spontaneous speech (as exemplified by a corpus of American English telephone dialogues, a.k.a. SWITCHBOARD). Variation in pronunciation with respect to segmental identity and duration can be accounted for in terms of a juncture-accent model, in which position of the segment within the syllable (i.e., onset, nucleus, coda), in tandem with knowledge of the associated stress-accent pattern, is used to interpret the inherently ambiguous phonetic information contained in the acoustic signal. Many properties of pronunciation variation can be accounted for in terms of such a model, including: (1) the prevalence of coda deletion, (2) the mutability of vocalic identity and (3) the relative stability of syllable onsets. The melding of phonetic and prosodic features within the syllable provides for efficient and reliable linguistic information coding.