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 pa
- PDF / 1,823,807 Bytes
- 15 Pages / 439.37 x 666.142 pts Page_size
- 73 Downloads / 205 Views
Abstract 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.
1 Introduction Many real-world problems such as pattern recognition and decision-making, involve input information from several different sources; the evidence from each source alone can only provide a partial account of all available information. A decision based on information from a single source may be sub-optimal or incorrect, and it is often the convergence of evidence from various sources that provides an accurate and reliable result. Thus, the appropriate aggregation of information from different sources contributes crucially to the success of a solution to such problems. For certain applications it may be possible to create a comprehensive model involving all sources of information with respect to the overall classification or decision. However, for many other tasks a comprehensive model can be difficult to develop; it is often more practical to first perform an evaluation based on each information stream and then combine the results into a single overall result. This latter approach is often referred to as multiple-information aggregation. D. Ruan et al. (eds.), Intelligent Sensory Evaluation © Springer-Verlag Berlin Heidelberg 2004
386
S. Chang and S. Greenberg
Besides requiring less complex models, the multiple-information aggregation approach has several additional advantages over the single, comprehensive-mode approach in practical applications. Performing evaluation individually on each information source minimizes the potentially harmful impact of background noise derived from various sources. Moreover, separating the initial evaluation from the aggregation process provides for flexibility in using a different computing strategy for each information source; certain methods, such as pooling of subjective evaluations, possess an inherent structure similar to the multiple-information aggregation framework and thus may benefit from such an approach. Because of requirements associated with computation and interpretation, as well as the inherent uncertainty, the general class of multi-criteria decision-making and multi-attribute optimization problems lends itself well to soft-computing based techniques. Besides the fuzzy-integration-based aggregation described in this paper, many other soft computing approaches have been
Data Loading...
