| 摘要 |
PURPOSE: A method for recognizing a voice and detecting an end point is provided to search accurately a voice section by improving an end point detection function under environment including noise. CONSTITUTION: An input voice is divided in frame units, and particular vectors for voice recognition and end point detection are detected according to each frame when a voice is inputted. The voice and a non-voice are identified by comparing a threshold value of a parameter with the particular vector. The input voice is determined as the non-voice if the particular vector is less than the threshold value or the input voice is determined as the voice if the particular vector is larger than the threshold value. An end point detection algorithm is performed when the input voice is determined as the voice. The input voice is determined as the voice or the end point by performing the end point detection algorithm. A result is transferred to a recognizing device when the end point is detected by the end point detection algorithm. A cart-decision tree is used when the input voice is determined as the non-voice. The threshold value is updated by the cart-decision tree. The voice or the non-voice is determined by comparing the updated threshold value with the particular vector.
|
