Post-filter optimization for multichannel automotive speech enhancement

In an automotive environment, quality of speech communication using a hands-free equipment is often deteriorated by interfering car noise. In order to preserve the speech signal without car noise, a multichannel speech enhancement system including a beamformer and a post-filter can be applied. Since employing a beamformer alone is insufficient to substantially reducing the level of car noise, a post-filter has to be applied to provide further noise reduction, especially at low frequencies. In this thesis, two novel post-filter designs along with their optimization for different driving conditions are presented. The first post-filter design utilizes an adaptive smoothing factor for the power spectral density estimation as well as a hybrid noise coherence function. The hybrid noise coherence function is a mixture of the diffuse and the measured noise coherence functions for a specific driving condition. The second post-filter design applies a new multichannel decisiondirected aprioriSNR estimator based on both temporal and spatial smoothing. For different driving conditions, both post-filters are instrumentally optimized: For the first post-filter, the optimal adaptive smoothing factor and the optimal hybrid noise coherence function are obtained. For the second post-filter, the weighting factors of the temporal and spatial smoothing parts are optimized. Compared to state-of-the-art post-filters, both post-filter designs employing the optimized parameters improve the overall noise reduction performance significantly for different driving conditions. Generally, manually finding the optimal parameterization of a noise reduction algorithm is a time-consuming task. In this thesis, the two new post-filter designs are thus instrumentally optimized by using a figure of merit (FoM). We define the FoM as an entity, which comprises three independent instrumental measures for the speech component quality, the level of noise attenuation, and the amount of musical tones. Particularly, a new weighted log kurtosis ratio measure is proposed for instrumental musical tones assessment in a black-box test manner, which does not mandate any knowledge of internal variables of the noise reduction algorithm under test and can be applied to a wide range of noise reduction algorithms. Subjective listening tests reveal that the weighted log kurtosis ratio measurements can provide a high correlation to the perceived amount of musical tones. In addition, a single-channel application example of jointly optimizing the smoothing factor and the aprioriSNR floor of the decision-directed aprioriSNR estimation is shown using an FoM. For some noise reduction algorithms, yet unknown optimal values of the parameters of interest are identified by applying the FoM-based instrumental optimization method and subjectively verified.