Singing Transcription from Multimedia Audio Signals using Contour Filtering

Abstract

With the accelerated development of multimedia technology, audio signals gradually occupy extremely large components. We propose a contour filtering model based on audio signals analysis by using the concept of salience function, and introduce the edge detection method in image processing to search out all the contours, which really realizes the analysis of the characteristics of the melody contours. The unique method combining audio signals with image analysis provides a more efficient analysis platform for singing signals. Experimental analysis using Music Information Retrieval Evaluation Exchange (MIREX) datasets show that based on human auditory effect achieve great results for multimedia audio signals and has more robust performance on complicated scene.