Application of Sound Recognition Techniques for Identification of the Squeak and Rattle Noises

Automatic classification and identification of squeak and rattle (S&R) issues from their sound is a computing tool of increasing productivity in automotive. This paper aims to investigate the usage of standard tools of music and speech recognition in application to S&R sounds in automotive which are usually noises-like and may include strong temporal domain signatures. Acoustic features of the sounds are extracted from audio samples into spectrogram-type data: the Mel-frequency cepstral coefficients (MFCC) features in non-stationary implementation which describe the audio spectral shape. Then, the extracted features are processed with two artificial intelligence techniques. These are Gaussian mixture model (GMM) and Support vector machine (SVM). As the results, the comparison of robustness of GMM and SVM methods is presented. Results show more than 50% and around about 45% recognition rates using mel frequency cepstral coefficients with SVM and GMM respectively.