Robust audio surveillance using spectrogram image texture feature

A sound signal produces a unique texture which can be visualized using a spectrogram image and analyzed for automatic sound recognition. In this paper, we explore the use of a well-known image texture analysis technique called the gray-level co-occurrence matrix (GLCM) for sound recognition in an audio surveillance application. The GLCM captures the distribution of co-occurring values at a given offset. Unlike most other similar research which derive features from the GLCM, we use the matrix values itself to form the feature vector with analysis carried out in subbands. When compared to a baseline feature from related work, the proposed spectrogram image texture feature (SITF) gives marginally lower results under clean and high signal-to-noise ratio (SNR) conditions but significantly better results are achieved at low SNR, where the baseline feature was seen to be less effective.