Progressive-Search Algorithms for Large-Vocabulary Speech Recognition

PROGRESSIVE-SEARCH ALGORITHMS FOR LARGE-VOCABULARY SPEECH RECOGNITION Hy Murveit John Butzberger Vassilios Digalakis Mitch Weintraub SRI International ABSTRACT We describe a technique we call Progressive Search which is useful for developing and implementing speech recognition systems with high computational requirements. The scheme iteratively uses more and more complex recognition schemes, where each iteration constrains the search space of the next. An algorithm, the Forward-Backward Word-Life Algorithm, is described. It can generate a word lattice in a progressive search that would be used as a language model embedded in a succeeding recognition pass to reduce computation requirements. We show that speed-ups of more than an order of magnitude are achievable with only minor costs in accuracy. 1. INTRODUCTION Many advanced speech recognition techniques cannot be developed or used in practical speech recognition systems because of their extreme computational requirements. Simpler speech recognition techniques can be used to recognize speech in reasonable time, but they compromise word recognition accuracy. In this paper we aim to improve the speed/accuracy trade-off in speeeh recognition systems using progressive search techniques. We define progressive search techniques as those which can be used to efficiently implement other, computationally burdensome techniques. They use results of a simple and fast speech recognition technique to constrain the search space of a following more accurate but slower running technique. This may be done iteratively---each progressive search pass uses a previous pass' constraints to run more ettieiently, and provides more constraints for subsequent passes. We will refer to the faster speech recognition techniques as "earlier-pass techniques", and the slower more accurate techniques as "advanced techniques." Constraining the costly advanced techniques in this way can make them run significantly faster without significant loss in accuracy. The key notions in progressive search techniques are: 1. An early-pass speech recognition phase builds a lattice, which contains all the likely recognition unit strings (e.g. word sequences) given the techniques used in that recognition pass. 2. A subsequent pass uses this lattice as a grammar that constrains the search space of an advanced technique (e.g., only the word sequences contained in a word lattice of pass p would be considered in pass p+l). Allowing a sufficient breadth of lattice entries should allow later passes to recover the correct word sequence, while ruling out very unlikely sequences, thus achieving high accuracy and high speed speech recognition. 2. PRIOR ART There are three important categories of techniques that aim to solve problems similar to the ones the progressive search techniques target. 2.1. Fast-Match Techniques Fast-match techniques[l] are similar to progressive search in that a coarse match is used to constrain a more advanced computationally burdensome algorithm. The fast match, however, simply uses the local speech signal to constrain the costly advanced technique. Since the advanced techniques may take advantage of non-local data, the accuracy of a fast- match is limited and will ultimately limit the overall technique's performance. Techniques such as progressive search can bnng more global knowledge to bear when generating constraints, and, thus, more effectively speed up the …

Content truncated to 500 words