Study of the Design and Implementation of Speech Keyword Recognition System based on Streaming Media

摘要

The paper mainly discusses the speech keyword recognition system dealing with the audio streaming media. With the help of the Microsoft Windows Media Format SDK(WMFSDK), a powerful front-end interface module is designed to extract audio stream from different streaming media and convert it to the audio format supported by the speech-recognizer. In order to rapidly spot keywords and reject out-of-vocabulary (OOV) words, the keyword-spotting strategy is put forward based on on-line garbage models. Studies show that this strategy works well in utterance verification. On the utterance verification stage, mixed with multi-confident measures, three classifiers are designed and compared, with the test results proved by different classifiers, the support vector machine (SVM) method is proved superior in performance to the Fisher and neural network(NN) method.