Single channel speech enhancement in the presence of additive noise

In this thesis, the performance of a short-time noise reduction method prior to an analysis/synthesis system based on a peak-picking algorithm has been analyzed for noisy speech signals, which have been degraded by additive white Gaussian noise at different signal-to-noise ratios (SNR) and different types of additive background noise. The reverberation effect is usually encountered in single-microphone speech enhancement methods, based on a SNR dependent spectral gain, and it limits the noise reduction performance of the system. Although, two-step noise reduction (TSNR) method solves this problem, when difficulty is encountered in estimating noise power spectral density (PSD), it causes some harmonic distortion in the enhanced speech. To overcome this problem, prior to an analysis/synthesis system, harmonic regeneration noise reduction (HRNR) method has been used. Analysis/synthesis system has been used because, by the use of peak-picking algorithm, high energy regions are represented, which causes a significant amount of data reduction, providing a simpler description of a speech signal and facilitating its further manipulation. In this thesis, speech signal, degraded by additive broadband noise at different SNR and different types of background noise has been enhanced by first TSNR method, then HRNR method and objective test results have been provided to compare the results. At the end of the noise reduction process it has been seen that, some types of noise could not be reduced due to inefficient noise estimation. The noise that could not be reduced at that stage has been reduced as a result of the enhancement procedure at analysis/synthesis system. Objective test results have shown the amount of enhancement in the noisy speech. All the applications have been performed by using MATLAB simulation program.