Surveillance Audio Attention Model Based on Spatial Audio Cues

摘要

For stereo audio surveillance in complex environment, we proposed a bottom-up audio attention model based on spatial audio cues analysis, and an environment adaptive normalization method. The traditional audio attention models are based on mono audio characters, such as energy, energy peak, or pitch. Their performance is limited by neglecting the spatial information. The spatial cues in audio stream provide additional information for attention analysis. And the dynamic updated background sound can help to reduce the environment effect. The preliminary experiment showed that the proposed model is an effective way to analyzing attention events, which is caused by rapid moving sound source, in stereo audio stream.