Low-complexity dangerous sound scene distinguishing method based on short-time energy and Mel-frequency cepstral coefficient combined novel vector quantization

Abstract

The invention belongs to the field of intelligent application type sound field distinguishing, and in particular relates to a dangerous sound scene distinguishing method based on short-time energy andMel-frequency cepstral coefficient combined vector quantization. The method specifically comprises the following steps: establishing a dangerous sound scene sound bank, constructing audio frequency and time frequency composite feature parameters, and introducing an improved vector quantization model for training the audio frequency feature parameters; and carrying out distortion degree matching by adopting a Euclidean distance measurement algorithm at the dangerous sound scene matching stage, and finding out the minimal average error distortion degree as the optimal matching, so that the recognizing effect is achieved.

Description

technical field [0001] The invention belongs to the field of intelligent application sound field discrimination, and in particular relates to a dangerous sound scene discrimination method combined with short-term energy and Mel cepstrum coefficient vector quantization. Background technique [0002] The realistic sound scene contains a wealth of useful information. When there is a specific potential danger in an environment, the detection effect of the dangerous sound environment can be achieved by identifying the key features of the sound in the scene. At present, the safety of children has aroused the great attention of the society and the majority of parents. As we all know, in the environment of children's active life, when some unconventional sounds appear, it means that there is potential danger in this environment. At this time, if the children and their parents can be reminded in time, the occurrence of dangerous incidents against children can be effectively avoided....

AI Technical Summary

Problems solved by technology

However, the current battery development has encountered a bottleneck. The battery capacity of existing wearable electronic devices is small and the battery life is relatively short. This puts forward higher requirements for the algorithms embedded in such devices. It is hoped that the embedded abnormal sound recognition algorithm should be simple and efficient. Computational complexity is lower

However, the complexity of the existing sound field recognition algorithms is relatively high, and the long recognition time cannot meet the requirements of portable electronic devices.

Images