Investigation of Improved Masking Noise for the Speech Privacy

Many sound masking products for the purpose of achieving speech privacy have emerged on the market in recent years. Over the time, quality of the masking noise has been improved a lot and still is studied for a better method. This thesis introduces a new way of generating masking noise which outperforms the commonly used methods. Using Adaptive Linear Predictive Coding (LPC) we can derive the model for various target sound and create a new masking signal which has exactly the same spectral envelope of the target sound and different phase. Two important aspects in evaluating the performance quality of the masking noise are masking capability and pleasantness. By testing these criteria in objective and subjective ways, we show superior performance of the proposed adaptive method in comparison with several existing techniques.