Real-time noise filtering with adaptive filters in heavy equipment soundscape

Abstract

In this master’s thesis, adaptive filters are used to abate the engine noise of heavy equipment and the changes in the soundscape are studied. The main objective of this work is to enhance the sound quality of human speech and shouting. The results are evaluated both with subjective tests and computationally. The first method used in achieving the goal is the Butterworth band-pass-filter, which is designed to preserve the frequencies possibly containing human speech and filter the rest of the frequencies. The resulting signal of the Butterworth filter is filtered with the adaptive filter targeting the engine noise. In this study two different adaptive filters are used, the Wiener filter and NLMS filter, and their performance is compared. In addition to the method of implementation, these filters also differ in that the Wiener filter does not use a reference signal for adaptation, but the noise is estimated from the input signal itself, while the NLMS filter uses a microphone in the engine compartment as a reference signal. The filtering system developed in this study is implemented on a developing platform, which is designed to be used by the end user, in other words, the operator of the heavy equipment. This is the reason why the results of the subjective tests are in focus in this study. According to both objective and subjective evaluations in this study, the engine noise deteriorates considerably and the speech coming outside the vehicle is cleaner. In the thesis the results were evaluated both with Sandvik Pantera DPI series drilling machine and a large diesel engine car, Nissan Pathfinder. The subjective evaluation is compared with the signal-to-noise ratio and signal-distortion ratio, which both indicated that the enhancement was successful. Even though the test conditions were not optimal, the results show that the adaptive filters can be used efficiently to filter the engine noise in real-time.