Audio Enancement using Nonlinear Time-Frequency Filtering

Forensic audio recordings may contain undesired noise that can impair source identification, speech recognition, and other audio processing requirements. In this paper several custom analysis/synthesis algorithms are presented based on a time-varying spectral representation of the noisy signal. The enhancement process adapts to the instantaneous signal behavior and alters the noisy signal so that the enhanced output signal is higher in quality than the unprocessed noisy input signal. Nonlinear and time-varying filters operate on the spectral representation in order to retain features that are attributable to the desired signal, such as human speech, while removing the features that are more likely to be due to the noise contamination.