Acoustic Effect Classification Based On Hidden Markov Model And Back-propagation(BP) Neural Network

Multimedia draws much attention from public recently due to the development of multimedia technology and the improvement of computation efficiency.Nowadays,appropriate sound clips for sound editing in broadcasting and television industry are searched only by auditory judgement,which is time-consumed and inefficient.This is because the audio asset is huge which often contain multiple sound sources also with abundantly semantic and auditory characteristics.Therefore,an automatic sound effect classifier is needed urgently?This study is first time in the area of broadcasting and television industry to investigate whether and how machine learning techniques are able to classify sound effect materials.The study firstly extracts the characteristic parameters of sounds for the database.Two different algorithms are used as a classifier.Sound clips are distinguished according to their similarity,also being identified by its sort.Finally,a sound classification protosystem based on BPNN is established.The core contributions of this study are1.Three characteristic parameters namely the short-time energy,short-time average zero-crossing rate and Mel frequency cestrum coefficients and its differences are extracted from 4704 audio fragments.Corresponding standard sample dataset is being established according to different types of algorithms in the experiment2.Two popular algorithms,including back-propagation neural network(BPNN)and Hidden Markov Model(HMM)are being used to train and test the samples respectively.The two algorithms performance is being tested as well.The study discusses the two algorithms also with the related research in the aspect of the algorithm structure,training time and recognition rate.The results show that the sound materials with confusing elements from complex sound source are classified easier based on BPNN training method,the average recognition rate is around 90%3.The sound classification protosystem based on BPNN is in order to help the sound mixer in the future.