Reduction of wind induced microphone noise using singular spectrum analysis technique

Wind induced noise in microphone signals is one of the major concerns of outdoor acoustic signal acquisition. It affects many field measurement and audio recording scenarios. Filtering such noise is known to be difficult due to its broadband and time varying nature. This thesis is presented in the context of handling microphone signals acquired outdoor for acoustic sensing and environmental noise monitoring or soundscapes sampling.Thethesis presents a new approach to wind noise problem. Instead of filtering, a separation technique is developed. Signals are separated into wanted sounds of specific interest and wind noise based on the statistical feature of wind noise. The new technique is based on the Singular Spectrum Analysis methodwhich has recently seen many successful paradigms in the separation of biomedical signals, e.g., separating heart soundfrom lung noise. It has also been successfully implemented to de-noise signals in various applications. The thesis set out with particular emphasison investigating the factor that determines and improves the separability towards obtaining satisfactory results in terms of separating wind noise components out from noisy acoustic signals. A systematicapproach has been established and developed within the framework of singular spectral separation of acoustic signals contaminated by wind noise. This approach, which utilisesa conceptual framework, has, in its final form, three key objectives; grouping, reconstruction and separability. This approach is offered through introducing new mathematical models particularly for window length optimisation along with new descriptive figures.The research question has therefore been addressed considering developing algorithms according to updated requirements from method justification to verification and validation of the developed system. This thesis follows suitable testing criteria by conducting several experiments and a case-study design, with in-depth analysis of the results using visual tools of the method and related techniques.For system verification, an empirical study using testing signals thatintroduces a large number of experiments has been conducted. Empirical study with real-world sounds has been introduced next in system validation phase after rigorously selecting and preparing the dataset whichis drawn from two main sources: freefield1010 dataset, internet-based Freesound recordings. Results show that microphone wind noise is separable in the singular spectrum domain after validating and critically evaluating the developed system objectively. The findings indicate the effectiveness of the developed grouping and reconstruction techniques with significant improvement in the separability evidenced by w-correlation matrix.The developed method might be generalised to other outdoor sound acquisition applications