Large-vocabulary dictation using SRI's DECIPHER speech recognition system: progressive search techniques

Survey of the State of the Art in Human Language Technology

“…Each pass prepares information for the next one, reducing the size of the search space. Details of these methods can be found in Alleva, Huang, et al (1993); Aubert, Dugast, et al (1994); Murveit, Butzberger, et al (1993); Kubala, Anastasakos, et al (1994).…”

“…The construction of the word lattice may involve only the execution of a Viterbi beam search with memorization of word scoring and localization, as in Aubert, Dugast, et al (1994), or may itself require multiple steps, as in Alleva, Huang, et al (1993); Murveit, Butzberger, et al (1993); Kubala, Anastasakos, et al (1994). Since the word lattice is only an intermediate result, to be inspected by other detailed methods, its generation is performed with a bigram language model, and often with simplified acoustic models.…”

“…In cases where the input sentence is uncertain, such as speech recognition, we may further generalize the above picture to a simultaneous search of the configuration space for the grammar and of a space of sentence hypotheses, represented for instance as a word lattice (Murveit, Butzberger, et al, 1993).…”

“…Lattice rescoring may require new calculations of HMM probabilities (Murveit, Butzberger, et al, 1993), may proceed on the basis of precomputed probabilities only (Aubert, Dugast, et al, 1994;Alleva, Huang, et al, 1993), or even exploit acoustic models which are not HMMs (Kubala, Anastasakos, et al, 1994). In Alleva, Huang, et al (1993), the last step is based on an ¡ search (Nilsson, 1971) on the word lattice, allowing the application of a long distance language model, i.e., a model where the probability of a word may not only depend on its immediate predecessor.…”

Handling compound nouns in a Swedish speech-understanding system

“…The system is capable of translating spoken utterances from the Air Travel Planning (ATIS) domain from Swedish into English, using a vocabulary of about 1 500 words. Continuous speaker-independent speech recognition is performed by a Swedish version of the DECI-PHER (TM) recognizer [9], and language processing is provided by a Swedish version of the SRI Core Language Engine [3,6].…”