Research on Violin Audio Feature Recognition Based on Mel-frequency Cepstral Coefficient-based Feature Parameter Extraction

Abstract

This paper focuses on the feature recognition of violin audio. After introducing the preprocessing method, the common feature parameters, linear predictive cepstral coefficient (LPCC) and mel-frequency cepstral coefficient (MFCC), were explained. Then, MFCC + △MFCC was used as the feature parameter. The parameters of support vector machine (SVM) were optimized using the firefly algorithm (FA). The FA-SVM method was used to recognize different violin audios. It was found that the identification rate of the FA-SVM approach was above 95% for different violin notes. The recognition effect was better when using MFCC + △MFCC as the feature parameter compared with LPCC and MFCC. The FA-SVM method achieved the highest recognition rate of 97.42%. The results demonstrate the reliability of the FA-SVM method based on MFCC feature parameter extraction. This method can be applied in practical audio recognition.