Second International Conference on Spoken Language Processing (ICSLP'92)
Banff, Alberta, Canada
In this paper, we describe a preliminary investigation of the use of syllabic fillers to model the out-of-vocabulary utterances or non-keywords in fluent speech for Spanish speech recognition systems. The Spanish language has a syllabic structure which suggest to use syllabic models to deal with the out-of-vocabulary utterances. Furthermore, Spanish demisyllable units has been used successfully as recognition units in the context of continuous speech recognition [7,8]. Thus, our proposal is to use a compact representation of the vocabulary words and the out-of-vocabulary words in terms of demisyllable units and syllabic fillers. The paper evaluates the performances of the syllabic fillers to deal with out-of-vocabulary utterances. As application task, we use the detection of digits in fluent speech.
Bibliographic reference. Lleida, Eduardo / Marino, Josť B. / Salavedra, J. / Bonafonte, Antonio (1992): "Syllabic fillers for Spanish HMM keyword spotting", In ICSLP-1992, 5-8.