作者: L. Bahl , R. Bakis , P. Cohen , A. Cole , F. Jelinek
DOI: 10.1109/ICASSP.1979.1170736
关键词:
摘要: The statistical training and decoding procedures developed at IBM Research can be used with a wide variety of acoustic processors. We have recently (July August 1978) achieved error-free or nearly results several different processors on sentences from the New Raleigh Language (vocabulary 250 words, perplexity 7.27 words). One these processors, which 0% error rate sentences, has been to decode without benefit syntactic guidance during process. On this much more difficult task, it an 8.8% word level, corresponding sentence 53%. All are non-segmenting produce output once every 10ms.