Joint dereverberation and noise reduction for binaural hearing aids and mobile phones

Speech signals captured by the microphones of a speech communication device are often distorted by interfering noise sources as well as room reverberation. Within this work, new algorithms and concepts enabling a joint reduction of room reverberation and environmental background noise for speech communication systems under adverse acoustic conditions are presented. The new strategies are based on measurements and recordings in realistic acoustical environments keeping in mind the specific acoustic conditions for the two main applications, i. e., binaural hearing aids and dual-microphone mobile phones. For binaural hearing aid applications, a new two-stage algorithm is proposed. It is based on a coherence model which takes the shadowing effects of the head into account as well as two combined estimators for the late reverberant speech Power Spectral Density (PSD) and background noise PSD. A key aspect is that all required acoustic parameters such as the Reverberation Time (RT) and Direct-to-Reverberation Energy Ratio (DRR) are estimated blindly from the noisy and reverberant input signals. The second algorithm, developed for dual-microphone mobile phones, exploits explicitly the Power Level Difference (PLD) of speech and all interfering sources and comprises a new noise PSD estimator and spectral weighting rule. The emerging technologies allow to jointly reduce early and late reverberation as well as additive background noise and are capable of improving the listening comfort and the speech intelligibility for hearing aid and mobile phone users.