Isolated Chinese lyrics with accompaniment recognition based on SVM

Abstract

The speech recognition technology is one of the hot spots in the field of audio technology. For the recognition of the lyrics with the accompaniment, there are two commonly used methods, one is applying automatic speech recognition technology to singing recognition, the other way is using sound classification, extracting audio features, and then using pattern matching classifier for classification. In this paper, we use sound classification method, adopt self-built experimental database where 31 classes Chinese isolated lyrics (Total 4650) are intercepted from different songs. And then use these words as the units. Considering speaking and singing sharing similar mechanism, we extract 39-dimensional MFCC feature parameters which are widely used in speech recognition. Combined with training materials, adjust kernel parameters and choose functions to train SVM classifier. After that, the trained SVM classification system is used to recognize the lyrics, and the average recognition accuracy rate is 42.80%.