| 摘要 |
PURPOSE: A speech recognition method is to enhance a speech recognition rate by assigning a state value proportional to the number of average analysis frames of an input word and sampling the state value. CONSTITUTION: Speech is recognized by sampling a state value according to an input word. When the state value is sampled, the input word is partitioned based on the number of frames according to an average analysis time. The average analysis time is ranged from 20 milliseconds to 30 milliseconds. The state value corresponding to half the number of the frames is assigned. Compared with a speech recognition method based on one word or a phoneme, the speech is recognized accurately by partitioning a pronunciation time. In case where there is no noise, the speech recognition method obtains a speech recognition rate of 96.09.
|
