VOICE RECOGNIZING APPARATUS AND METHOD USING NEURAL NETWORK

摘要

PURPOSE: A voice recognizing apparatus and method using a neural network are provided to improve a voice recognizing performance by applying a feature vector normalization and a neural network to a voice recognizing technique. CONSTITUTION: A voice recognizer(110) receives an analog voice signal through a microphone and converts it into a digital signal. A voice section subdivision section(120) subdivides the digital voice signal in units of a frame. A feature vector extractor(130) extracts a feature vector from the digital voice signal in units of a frame from voice section subdivision section(120). A feature coefficient normalizing section(140) converts a feature vector having a various length into a feature vector having a set length. A neural network algorithm section(150) learns a feature vector having a predetermined length and a neural network coefficient learned in words to output a updated neural network coefficient. A neural network coefficient supply section(160) supplies the updated neural network coefficient. A detection result output section(170) outputs a detecting result corresponding to the updated neural network coefficient from neural network coefficient supply section(160) to an external device.