Speech recognition method using statistical probability calculations

摘要

The speech recognition method involves using an acoustic analyser (1), an identification element (2), an acoustic model register (3) and a speech model register (4) with a grammar control element (5). The analyser provides digital signals as a time series of parameters such as the cepstrum and delta cepstrum values. The identification stage calculates the probability of the speech being in agreement with the acoustic model register and the speech model register. The method calculates the statistical probability series from the acoustic model and identifies the recognition result. Then only one record is displayed. Preferably the controller also rules out any mistakes in grammar.