Speech detection method in complex and strong noise environment

Abstract

The invention belongs to a voice detection method in a complex and strong noise environment represented by shortwave channels, including extracting sample data, preprocessing sample data, extracting signal features, training weak hypotheses according to the balance of sample data, and establishing voice The detection model detects the voice real-time signal data stream. The invention improves the timing of training exit in the background technology and the excessive distortion of data samples caused by the imbalance of data categories in the training process. At the same time, it combines the conventional spectral subtraction method to process broadband noise, increasing the speech and noise, and doubled the feature dimension; therefore, more effective weak hypotheses can be integrated into the classification model, effectively improving the performance of the classification model, and the detection accuracy can be increased by 10% compared with the background technology above. Therefore, it has the characteristics of strong ability to detect short-wave speech in a complex and strong noise environment, and high detection accuracy.

Description

Speech detection method in complex strong noise environment Technical field [0001] The present invention relates to a method for detecting speech in complex and strong noise environments represented by shortwave channels. Background technique [0002] The voice detection method is one of the key technologies and core contents necessary for voice reconnaissance. Accurate voice detection The level of accuracy directly determines the reconnaissance ability of voice intelligence. Most of the traditional speech detection researches are aimed at speech endpoint detection. The test is to detect the signal segment containing speech from the continuously input signal, that is, to make speech / non-speech judgment, the purpose is To find the exact start and end points of each speech segment in the signal. In Speech Detection: Adaptive Enhancement Methods (English as AdaBoost) is one of the most popular and successful ensemble learning (model building) methods, since its birth, ma...

AI Technical Summary

Problems solved by technology

Recording a large number of pure noise signals consumes a large amount of storage resources, wastes manpower, and increases the difficulty of subsequent processing.

degree, seriously reducing the efficiency of follow-up processing and other disadvantages