Cepstrum-Based Preprocessing for Howling Detection in Speech Applications
Conventional howling detection algorithms exhibit dramatic performance degradations in the presence of harmonic components of speech that have the similar properties with the howling components. To solve this problem, this paper proposes a cepstrum preprocessing-based howling detection algorithm. First, the impact of howling components on cepstral coefficients is studied in both theory and simulation. Second, according to the theoretical results, the cepstrum pre-processing-based howling detection algorithm is proposed. The Receiver Operating Characteristic (ROC) simulation results indicate that the proposed algorithm can increase the detection probability at the same false alarm rate. Objective measurements, such as Speech Distortion (SD) and Maximum Stable Gain (MSG), further confirm the validity of the proposed algorithm.
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.