Effect of Speaker-Microphone Proximity on Pop Noise: Continuous Wavelet Transform-Based Approach

The problem of Voice Liveness Detection (VLD) has been explored recently, wherein pop noise is used as the discriminating acoustic cue between live and spoofed speech. In this work, we propose an algorithm to capture pop noise energy using Bump wavelet, wherein we test our proposed algorithm to analyse the effect of distance (between the speaker and the microphone) on the strength of pop noise captured by the microphone. Furthermore, the type of phoneme uttered also plays an important role, when dealing with pop noise. Therefore, in this work, we have analyzed the variability of pop noise strength with distance, for each of the phoneme types. To the best of the authors’ knowledge and belief, this is the first work of its kind to take into account the distance as well as the phoneme type for the analysis of pop noise event in the speech signal. We have compared our proposed CWT-based approach with STFT-based approach using trendline equations. Our analysis shows that CWT-based approach is better suited for VLD, and the strength of the pop noise for phoneme classes, such as liquids and nasals is the least affected by distance. On the other hand, pop noise due to plosives and fricatives is the most affected by distance variability.