A B C D E F G H I J K L M N O P Q R S T U V W X Y Z All
Kokate, V. K.
- Improving Performance of Multiclass Audio Classification Using SVM
Authors
1 Department of Electronics and Telecomm, College of Engineering, Pune, IN
2 College of Engineering, Pune, IN
3 Electronics and Telecommunication Department, IN
Source
Data Mining and Knowledge Engineering, Vol 2, No 5 (2010), Pagination: 95-103Abstract
Audio classification has found widespread use in many emerging applications. It involves extraction of vital temporal, spectral and statistical features, and using these in creating an efficient classifier. Most of the audio classification work has been done on binary class classification. In our work we suggest best suited features for classification of different audio classes. Here, we present an algorithm for audio classification that is capable of segmenting and classifying an audio stream into speech male, speech female, music, noise and silence. The speech clips are further segment into voiced and unvoiced frames. A number of timbre features have been discussed, which distinguish the different audio formats. For pre classification, Probability Density Function (PDF), which is a threshold-based method, is performed over each audio clip. For further classification, K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) Classifiers are proposed. Experiments have been performed to determine the best features of each binary class. Utilization of these features in multiclass classification yielded accuracy 96.34% in audio discrimination.
Keywords
Audio Feature Extraction, Bayesian Classification, K-Nearest Neighbor, Support Vector Machine.- Automatic Identification of Tabla Tempo and Transcription of Bols
Authors
1 Department of Electronics and Telecom, College of Engineering, Pune, IN
2 College of Engineering, Pune, IN
3 Electronics and Telecom Department, College of Engineering, Pune, IN
Source
Artificial Intelligent Systems and Machine Learning, Vol 2, No 5 (2010), Pagination: 72-78Abstract
Automatically extracting music information is gaining importance as a way to structure and organize the increasingly large number of music files digitally available and has become an important part of multimedia research. This research becomes more interesting as well as challenging when the music analysis is done according to the instruments used in the making of music. One such very popular percussion instrument called Tabla widely used in accompanying Indian Classical music recitals is analyzed based on the collection of digital recordings. The database consists of popular taals commonly used in Indian classical music. Bols are the basic notes of Tabla. A taal is a predefined sequence of Tabla bols. Different such bol arrangements give rise to various taals. Based on the speed of repetition of bols, the taals are broadly classified into low (Vilambit), medium (Madhya) and fast (Drut) tempo. Thus a tempo represents the rhythmic information of a taal.In this paper, an automatic system for identifying and transcribing Tabla bols of different tempos is explored. The transcription process is based on three main steps: firstly tempo of the audio clip is identified using autocorrelation technique. Secondly, the recorded clip is segmented where each segment represents a bol. In the third step, Mel-Frequency Cepstral Coefficients (MFCC) features are extracted from the separated bols to form templates for pattern classification. Two pattern classification techniques namely Dynamic Time Warping (DTW) and Vector Quantization (VQ) are analyzed and compared for evaluating the performance of bol identification. Overall bol identification accuracy of the system for tempo independent classification is 96.18%.