The audio mainly realizes the following functions: A, extracting the envelope of the audio, realizing the volume is too large, too small, and the intercept detection function; B, comparing the energy values before and after the audio break, to realize the sound detection function; C, the audio First-order straight line fitting processing is performed on the volume value at the beginning and the end to obtain the slope and vertical offset of the fitted straight line, so as to realize the detection function of the first-end fade-in and fade-out effect; D. The typical current sound spectrum has obvious bright line features. By calculating the audio energy and the corresponding variance value, this characteristic is analyzed to realize the function of current sound detection. The invention has the advantages that the method for extracting the envelope is adopted, and the accuracy of the volume detection is further improved; the detection of the fade in and out is well performed, and the accuracy is high; the method has a good detection effect on the specific current sound.
With the continuous improvement of the quality of life of modern people, people's pursuit of music quality is becoming more and more urgent. High-quality MP3s, lossless format music, and the like, music appreciation sites and software are also increasingly involved in people's daily lives. However, in a large number of music music libraries, the quality of the audio is not uniform, and the manual detection of the sound quality requires a huge amount of work and is not sustainable. Therefore, this requires a good measure to solve this problem.
The method of audio feature detection, these features will affect the auditory effect of the audience to a certain extent, thus providing a certain reference for the artificial detection of sound quality. Through this method, the audio features can be automatically detected, the huge workload and manpower risk of manual detection are reduced, and the detection efficiency is effectively improved. At the same time, the characteristic false detection rate caused by human factors can be reduced, and the accuracy of the detection is further ensured.
The extraction of audio features, then analyzes the effectiveness and robustness of these features to distinguish different audio types from two aspects. Finally, based on these features, a support vector machine (SVM) classifier is used to train and test on an audio dataset of approximately 5 hours, dividing the audio into five predefined categories: mute, pure speech, impure speech, music, and Ambient sound. Experimental results show that using effective audio features, it is possible to classify audio scenes into different types.
