In this paper a new, tree-trellis based fast search for finding the N best sentence hypotheses in continuous speech recognition is proposed. The search consists of two parts: a forward, time-synchronous, trellis search and a backward, time asynchronous, tree search. In the first module the well known Viterbi algorithm is used for finding the best hypothesis and for preparing a map of all partial paths scores time synchronously. In the second module a tree search is used to grow partial paths backward and time asynchronously. Each partial path in the backward tree search is rank ordered in a stack by the corresponding full path score, which is computed by adding the partial path score with the best possible score of the remaining path obtained from the trellis path map. In each path growing cycle, the current best partial path, which is at the top of the stack, is extended by one arc (word). The new tree-trellis search is different from the traditional time synchronous Viterbi search in its ability for finding not just the best but the N-best paths of different word content. The new search is also different from the A * algorithm, or the stack algorithm, in its capability for providing an exact, full path score estimate of any given partial (i.e., incomplete) path before its completion. When compared with the best candidate Viterbi search, the search complexities for finding the N-best strings are rather low, i.e., only a fraction more computation is needed.

A lot of work remains to be done in the domain of a better integration of speech recognition and language processing systems. This paper gives an overview of several strategies for integrating linguistic models into speech understanding systems and investigates several ways of producing sets of hypotheses that include more "semantic" variability than usual language models. The main goal is to present and demonstrate by actual experiments that sequential coupling may be efficiently achieved by word-lattice syntactic analyzers, efficiently parsing the huge number of hypothesis (i.e. possible sentences) contained in the lattice produced by the speech recognizer. 1. Motivations The past decade has seen significant progress in speech recognition technology: word (recognition) error rates continue to drop by a factor of 2 every two years (Rabiner et al., 1996) and high performance systems are now becoming available. Several factors have contributed to this rapid progress: ffl Generalisati...

In this paper, weintroduce an e#cient algorithm for the exhaustive search of N best sentence hypotheses in a word graph. The search procedure is based on a two-pass algorithm. In the #rst pass, a word graph is constructed with standard time-synchronous beam search. The actual extraction of N best word sequences from the word graph takes place during the second pass.