摘要 |
PURPOSE: A method for recognizing emergency speech using a GMM(Gaussian mixture model) is provided to recognize an emergency in a CCTV environment which dynamically shows the emergency by detecting an emergency word by using GMM. CONSTITUTION: Noise is removed from an input speech signal. An initial point and an endpoint of the speech signal are detected. High frequency in the detected speech signal is emphasized. A feature vector of the detected speech signal is extracted based on a MFCC(mel-frequency cepstral coefficient). A global GMM is constructed by using an extracted feature vector. Emergency and nonemergency words are detected through a global GMM. The local GMM is constructed for recognition of a detected emergency word. The emergency word is recognized through the local GMM. |