Robust audio hashing based on discrete-wavelet-transform and non-negative matrix factorisation

Robust audio hashing defines a feature vector that characterises the audio signal, independent of content preserving manipulations, such as MP3 compression, amplitude boosting/cutting, hiss reduction, and so on. It provides a tool for fast and reliable identification of content in audio communications. In this study, the authors propose a new audio hashing based on discrete wavelet transform and non-negative matrix factorisation (NMF). The desirable property of NMF for hashing algorithm is its non-negative constraints, which result in bases that capture local feature of the audio, thereby significantly reducing misclassification. In addition, to ensure perceptual robustness, NMF is performed on the coarse wavelet coefficients, which are a low-pass approximation of the audio and not easy to change by content preserving manipulations. Experimental results over a large database reveal that the proposed scheme is more robust and provides much stronger discrimination than the conventional energy spectrum-based hashing algorithm, and that the proposed scheme can be applied in broadcast monitoring, successfully.