Audio Signals Recovery With Time-Frequency Estimation Using Short-Time Fourier Transform

Recovered audio signals contain serviceable information. In the context of audio signal analysis, the best trade-off between frequency estimation and time has to be made due to Heisenberg’s principle [2] which asserts that we cannot obtain the specific frequency, and time, t with the same amount of precision simultaneously. In this study, we use higher moments in attempting to reconstruct the time series data, and in finding an optimal bin size of the time window using short time Fourier transform. The performance of these two estimators, Doane and modified Doane, has been studied based on their mean squared errors (MSE).