期刊名称:Journal of Theoretical and Applied Information Technology
印刷版ISSN:1992-8645
电子版ISSN:1817-3195
出版年度:2014
卷号:70
期号:3
出版社:Journal of Theoretical and Applied
摘要:Audio mining is a technique by which the core of an audio signal can be automatically searched and analyzed. This research work addresses feature extraction from audio and audio similarity measures and proposes an algorithm to mine any type of audio data. This Hybrid Algorithm for Audio Mining (HAAM) consists of two different phases. The first phase named Training Phase, takes Training Audio data as input and then Sonogram, Spectrum Histogram, Periodicity Histogram and Fluctuation Pattern for the audio data are calculated. Then the features are extracted using Mel Frequency Cepstral Coefficient (MFCC) and the essential features are reduced from these for further processing. Then, classification is done by using a supervised classification technique. Then the features of the audio files are trained using Gradient Descent Adaptive Learning Back propagation Network. The second phase is the Testing Phase, which takes Testing Audio data as input and then preprocessing is done and the features are extracted. Then classification is done using the findings from the trained data by matching the features of the test audio with the equivalent or closest feature audio classes. The working prototype of this algorithm has been implemented and tested. During the experiments, Instrumental music data have been used as input and the system performed well and classified the input music in an accurate manner and presented the results obtained. The system is suitable for classifying the different types of audio data and can be used in many applications including speech recognition, audio classification in scientific research and engineering, audio data comparison and detection in audio surveillance applications.
关键词:Audio Mining; Gradient Descent Neural Network; K-Means Classifier; Mel Frequency Cepstrum Coefficient; Principal Component Analysis