Inverse Gabor transform for speech enhancement

In this paper, the Inverse Discrete Gabor Transform (IDGT) is proposed for signal recovery buried in board-band nonstationary noise. Time-frequency masking filtering technique is implemented to reject the noise from corrupted speech while at the same time maintaining the desired waveform. A synthetic multicomponent non-stationary test signal made up of two chirps was first used to simulate noise; the signals were then separated using this technique. Four English speech signals recorded in different environments such as airport, restaurant, and train buried in wide-band noise were reconstructed. The extracted signals were then compared with the original ones in terms of cleanness and noise removal. The implemented procedure is suitable for this type of wide-band non-stationary interference, which cannot be canceled in the Fourier (frequency) or time domain.