Real time speech recognition.

07 January 1986

New Image

The performance of speech recognition in real time, that is, computing a recognition choice in a time comparable to the length of the spoken utterance, requires special tailoring of speech recognition algorithms and the computing structures that execute them. For example, because real time speech recognition computations are performed while speech input is in progress, stream processing algorithms, which can only act on information based on the portion of speech that has already arrived, are emphasized over block processing techniques, which can postpone decisions concerning earlier portions of the waveform until the entire speech input is available.