Plan refinement in a knowledge-based system for automatic speech recognition

Frame-level phoneme classification using inductive inference

“…The next step is the development of feature detectors that can identify and extract these cues from the speech signal. The expert's decision-making process is encapsulated in a set of rules for classification (Lamel & Zue, 1984;Espy-Wilson, 1986De Mori & Lam, 1986;Zue & Lamel, 1986;Bulot & Nocera, 1989). The main problem here is that spectrogram-reading knowledge is tied with the ability of experts to see "events" in the spectrogram and determine the corresponding sounds or phones.…”

“…Still, the identification, development and formulation of the rules is performed manually by utilizing the expert. Thus, there is a strong reliance on the ability of the expert and the rule developer to identify and extract correctly the events in the speech pattern (Carbonell, et al, 1984;Carbonell, et al, 1986;Green & Wood, 1986;Mizoguchi, et al, 1986;De Mori & Lam, 1986;Stern, et al, 1986;Fohr et al, 1989). The main disadvantage of this technique is that we still need to formulate hundreds of rules, which are generally interrelated and very difficult to manage.…”