Learning Algorithms for the Cancellation of Old Recordings Noise
The elaborated software package for digital restoration of sound provides a system of digital operations applied to the registered signal of old records or magnetic tapes in order to bring the sound image back to its initial form. Particular attention was paid to the detection and elimination of scratches, hiss and similar types of impulse noise. A learning algorithm based on neutral network simulation was applied to this task. Another learning algorithm implementing the Wiener adaptive filtering procedures was used to continuous noise suppression. Applied methods and obtained experimental results are presented.
This paper costs $33 for non-members and is temporarily free for AES members.