The energy sources for magnetotellurics (MT) at frequencies above 8 Hz are electromagnetic waves generated by distant lightning storms propagating globally within the earth-ionosphere waveguide. The nature of the sources and properties of this waveguide display diurnal and seasonal variations that can cause significant signal amplitude attenuation, especially at 1-5 kHz frequencies - the so-called audiomagnetotelluric (AMT) dead band. This lack of energy results in unreliable MT response estimates; and, given that in crystalline environments ore bodies located at some 500-1000-m depth are sensed initially by AMT data within the dead band, this leads to poor inherent geometric resolution of target structures. We propose a new time-series processing technique that uses localization properties of the wavelet transform to select the most energetic events. Subsequently, two coherence thresholds and a series of robust weights are implemented to obtain the most reliable MT response estimates. Finally, errors are estimated using a nonparametric jackknife algorithm. We applied this algorithm to AMT data collected in northern Canada. These data were processed previously using traditional robust algorithms and using a telluric-telluric magnetotelluric (TTMT) technique. The results show a significant improvement in estimates for the AMT dead band and permit their quantitative interpretation.