Endpoint Detection Method Using Frequency Domain Information in Colored Noisy Environment

摘要

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a speech recognition technology, and more particularly, to a method of detecting an end point using information of a frequency domain in a colored noise environment for finding an end point in an isolated word speech recognition apparatus.;An initialization setting process of a counter value and an area indication indicating a state of a voice or non-voice area, a parameter extracting process for separating voice and non-voice after completion of initialization setting, a region comparison If the extracted parameter value is smaller than the threshold value in the non-speech area, the counter is operated. If the extracted parameter value is larger than the threshold value, the non-speech area is changed to the speech area. If the speech area is smaller than the threshold value, And a step of extracting the parameter from a next input signal if the area is changed to a non-speech area, and if the value is larger than the threshold value, Is compared with the non-speech setting value W is greater the smaller stop algorithm if characterized in that it comprises a counter comparison process of repeating a process of extracting a parameter from the signal input to the next. Therefore, it is possible to easily obtain the end point of the voice section in the color noise which changes with time or the color noise which is difficult to detect by the parameters such as the energy or the zero crossing rate.