Violin Note Classification

Although much research has been carried out on finding features for instrument recognition systems, little work has focused on specifically the violin's entire timbre space. Suitable features from which a computer can assess the quality of a violinist's playing have been sought and the classification of violin note sound quality is investigated in this paper. The eventual outcome of this work can be applied in various systems including the development of a violin or bowed string instrument teaching aid, in automatic music transcription and information retrieval or classification systems. gain better understanding of the relationship tween violin playing technique and the sound produced, a suitable means of quantifying and classifying these differences is needed. This is so that guidelines may be established for not only good violin sound but also for poorer or beginner violin playing with the ultimate aim of developing a computer based violin teaching aid, of which none exists. The more general area of quantifying beginner and professionals standard legato violin note samples using signal processing techniques was presented in (1, 2). This has enabled the representation of violin sounds by suitable descriptors. Violin playing faults have been identified and are limited to nine faults at this stage. This paper considers the classification of violin notes using up to fifteen features. Two tasks are put to a k- means nearest neighbour classifier: the first is the detection of beginner note from a professional standard note and the second is much more specific, involving individual fault detection. In the following sections, existing research is briefly presented, followed by a description of the data set requirements and how it was obtained, after which the listening tests are detailed. The choice and brief explanation of the features used in the classifier are then given, followed by the classification method and results obtained.