Signal Enhancement with Variable Span Linear Filters

In this book, the novel concept of variable span speech enhancement filters is introduced and it is shown how it can be used for noise reduction in various ways. The variable span filters combine the ideas of optimal linear filters with the ideas of subspace methods, as they involve the joint diagonalization of the correlation matrices of the desired signal and the noise. It is shown how some well-known filter designs, like the minimum distortion, maximum signal-to-noise ratio, Wiener, and tradeoff filters along with new generalization of these, can be obtained using the variable span filter framework. It is then shown how the variable span filters can be applied in various contexts, namely in single-channel STFT-based enhancement, in multichannel enhancement in both the time and STFT domains, and, finally, in time-domain binaural enhancement. In these contexts, the properties of these filters are analyzed in terms of noise reduction capabilities and desired signal distortion, and this analysis is validated and further explored in simulations. Furthermore, Matlab code will be included so that all the main ideas can be easily implemented by the readers.