Historic recordings usually have degraded audio quality because of their age, improper storage, and the shortcomings of the original media. One typical problem is the presence of impulsive disturbances. Recordings that suffer from clicks and crackles can be processed by impulse-restoration algorithms to improve their audio quality. This report presents a new algorithm that classifies one-second frames of an audio recording based on the existence of impulsive disturbances. The algorithm uses supervised learning. It is shown that existing impulse-restoration algorithms suffer from degradation of the desired signal if the input SNR is high and if no manual parameter adjustment is possible. This would make automatic restoration of large amounts of diverse archive audio material unfeasible. The proposed classification algorithm can be used as a supplement to an existing impulse-restoration algorithm to alleviate this drawback. An evaluation using a large number of test signals shows that high classification accuracy can be achieved, making automatic impulse restoration possible. Results show that prewhitening of the input signal by means of a phase-only transform serves to increase the detectability of disturbance impulses, which can also be used as a detection enhancement method for impulse-restoration algorithms.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=19356
(426KB)
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 Join the AES. If you need to check your member status, login to the Member Portal.
