Enhanced Automatic Noise Removal Platform for Broadcast, Forensic, and Mobile Applications
We present new enhancements and additions to our novel Adaptive/Automatic Wide Band Noise Removal (AWNR) algorithm proposed earlier. AWNR uses a novel framework employing dominant component subtraction followed by adaptive Kalman filtering and subsequent restoration of the dominant components. The model parameters for Kalman filtering are estimated utilizing a multi-component Signal Activity Detector (SAD) algorithm. The enhancements we present here include two enhancements to the core filtering algorithm, including the use of a multi-band filtering framework as well as a color noise model. In the first case it is shown how the openness of the filtered signal improves through the use of a two band structure with independent filtering. The use of color noise model on the other hand improves the level of filtering for wider types of noises. We also describe two other structural enhancements to the AWNR algorithm which allow it to better handle respectively dual microphone recording scenarios and forensic/restoration applications. Using an independent capture from a noise microphone the level of filtering is substantially increased. Furthermore for forensic applications a two/multiple pass filtering framework in which SAD profiles may be fine tuned using manual intervention are desirable.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is temporarily free for AES members.