Research On Wideband Speech Enhancement Algorithms

Single channel broadband speech enhancement is widely used in everyday life, and is also a difficult but challenging topic of signal processing .Although classical subtractive-type algorithms perform well in reducing the background noise, they also leave uncomfortable"musical noise"behind, especially in the case of low SNR environment. In order to achieve a better trade-off between the amount of noise reduction, the speech distortion, and the level of musical residual noise, the masking properties of the human hearing system are taken into consideration, thus the process of signal is constant with the human auditory system . This thesis examines both traditional noise suppression algorithms and two methods based on standard Johnston masking model. It is shown in both the objective and subjective tests that methods based on masking properties, which are modified and improved in this paper, provide speech of better clearness and intelligibility than conventional methods do, and it is computationally efficient as well. In addition, several end detection methods are also examined and tested.