Fused models for noise reduction in speech processing

We evaluate and compare two approaches for noise reduction in speech processing. One model is a recently developed implementation of a cochlea model that is solved in the time domain and that accounts for the influence of hair cells outside the inner ear membrane. The other approach involves statistical de-noising techniques based on wavelets. Our results indicate that the cochlea model achieves a stronger performance in removing noise, but requires significantly more computational effort. We propose the implementation of a fused model combining features of the cochlea model with wavelet-based techniques to achieve improved noise reduction properties at smaller computational costs.