A real-time training-free laughter detection system based on novel syllable segmentation and correlation methods

In this paper, a laughter detection system based on the correlation characteristic of signals is proposed. The advantages of the system are speaker independent, low-computational and training-free. To achieve the goal, a modified autocorrelation function (MACF) is combined with a new approach called vocal tract transfer detector (VTTD) for segmenting an input signal into a syllable stream. Next, based on each syllable´s Mel-scale frequency cepstral coefficients (MFCCs), the correlation between two consecutive syllables is measured by the dynamic time warping (DTW) algorithm. The consecutive syllables with high correlation are considered as a laughter segment. In our experimental result, the proposed system can achieve an accuracy rate of 88.67%. Besides, compared with the baseline, the proposed system can reduce the word error rate (WER) of syllable segmentation by 5.9%. Such results indicate that the proposed method is effective in detecting laughter, thereby demonstrating the feasibility of the system.