Research Of The Bad Audio Detection Based On MFCC

With the development of science and technology, the network and multimedia has become an indispensable part of people’s life, work and study, to bring convenience to people, but also provides a place for the bad culture communication, so the detection of video sickness crunch time, detection in order to achieve the network bad video accurately and rapidly, which gives people a healthy network. Audio is an important part of video. In order to improve the detection accuracy rate, this detection can better detect bad video’s bad audio part. This paper based on the bad video’s audio part to detect bad audio.In this paper, aiming at the bad video audio part, detection of bad audio, the audio content. But the bad video voice for isolated word, modal verbs and auxiliaries. Therefore the research of this paper is the speech audio detection based on sex.In the audio detection, mainly is to detect the speech signal from an audio signal, so to filter noise, music, the remaining pure voice and speech, audio classification, isolated environment, sound mute. That is the detection on the original audio signal, and completing the preliminary treatment.The extraction and selection of the extraction and the acoustic characteristics is an important part of speech recognition. In this paper, by comparing the characteristic coefficient, the final selection is MFCC coefficients, and the extraction method is improved, process description.In this paper, the audio detection method for pattern matching algorithm, in the training phase, will collect the bad audio sample template to the sample database, and its feature vector as the template in the template library. In the recognition stage, the input feature vector of the speech in a similarity with each template in the template library comparison, the output will be the highest similarity as the result of recognition.At the last introduces some algorithms, such as DTW, HMM, neural network. The direction and extension can be used as future research.Detection model of this study is the final audio for the pretreatment of audio, try to realize part of audio classification in the treatment process, and extracted MFCC coefficients are codebook formation of vector quantization based on LBG, finally, pattern matching, in order to realize the bad audio detection.The filtering and endpoint detection were used in the measurement system. This system adopts the method of combining audio classification technology and pattern recognition technology. The bad audio, animal sounds, music and normal speech four audio were as samples, and then the audio and the sample were compared. Compared with the most basic detection system, improves the detection rate was improved, but the detection time increased slightly. To ensure that within the scope of certain time, improve detection rate.