ResearchBib Share Your Research, Maximize Your Social Impacts
Sign for Notice Everyday Sign up >> Login

Automatic detection of audio defects using parallel computing

Journal: Software & Systems (Vol.35, No. 3)

Publication Date:

Authors : ; ;

Page : 428-437

Keywords : intel xeon phi; spectrogram; the anomaly; defect; audio record;

Source : Download Find it from : Google Scholarexternal

Abstract

The paper is devoted to research aimed at automatic detection of defects and anomalies that occur in the audio record digital signal. Defect detection methods are mainly used in digitizing analog audio records and restoring damaged signals. Anomaly detection methods have a wide range of applications including the development of security and environmental monitoring systems, the identification of artificially edited records, restoration of archival audio recordings of cultural value for a certain time peri-od of society development and formation, the fight against so-called deepfakes, encryption and decryption of classified information encoded in audio data and much more. Modern technologies and techniques make it possible to efficiently eliminate the found defects by mathematical manipulation of the signal by an audio engineer or using smart and adaptive digital signal editing tools. However, for this purpose, the defect must be accurately detected and localized, its type and the possible origin must also be determined. There is special software developed within the framework of this work to solve the problem of automatic detection of defects in a digital signal of an audio record. It was verified on digitized audio records of various quality. Since digital media data including audio records have a large size, the aspect of parallel distributed processing is of particular importance during the analysis. Due to this fact, the de-veloped defect detection code was upgraded to take into account the need to run on Intel Xeon Phi Knights Landing massively parallel microprocessors and demonstrated high scaling efficiency.

Last modified: 2023-02-10 17:48:45