Industrial-Grade CNN-Based System for the Discrimination of Music Versus Non-Music in Radio Broadcast Audio DOI Creative Commons
Valerio Cesarini,

Vincenzo Addati,

Giovanni Costantini

et al.

Information, Journal Year: 2025, Volume and Issue: 16(4), P. 288 - 288

Published: April 3, 2025

This paper addresses the issue of distinguishing commercially played songs from non-music audio in radio broadcasts, where automatic song identification systems are commonly employed for reporting purposes. Service call costs increase because these need to remain continuously active, even when music is not being broadcast. Our solution serves as a preliminary filter determine whether an segment constitutes “music” and thus warrants subsequent service identifier. We collected 139 h non-consecutive 5 s samples various labeling segments talk shows or advertisements “non-music”. implemented multiple data augmentation strategies, including FM-like pre-processing, trained custom Convolutional Neural Network, then built live inference platform capable monitoring web streams. was validated using 1360 newly samples, evaluating performance on both chunks 15 buffers. The system demonstrated consistently high previously unseen stations, achieving average accuracy 96% maximum 98.23%. intensive pre-processing contributed performances with benefit making inherently suitable FM radio. has been incorporated into commercial product currently utilized by Italian clients royalty calculation

Language: Английский

Industrial-Grade CNN-Based System for the Discrimination of Music Versus Non-Music in Radio Broadcast Audio DOI Creative Commons
Valerio Cesarini,

Vincenzo Addati,

Giovanni Costantini

et al.

Information, Journal Year: 2025, Volume and Issue: 16(4), P. 288 - 288

Published: April 3, 2025

This paper addresses the issue of distinguishing commercially played songs from non-music audio in radio broadcasts, where automatic song identification systems are commonly employed for reporting purposes. Service call costs increase because these need to remain continuously active, even when music is not being broadcast. Our solution serves as a preliminary filter determine whether an segment constitutes “music” and thus warrants subsequent service identifier. We collected 139 h non-consecutive 5 s samples various labeling segments talk shows or advertisements “non-music”. implemented multiple data augmentation strategies, including FM-like pre-processing, trained custom Convolutional Neural Network, then built live inference platform capable monitoring web streams. was validated using 1360 newly samples, evaluating performance on both chunks 15 buffers. The system demonstrated consistently high previously unseen stations, achieving average accuracy 96% maximum 98.23%. intensive pre-processing contributed performances with benefit making inherently suitable FM radio. has been incorporated into commercial product currently utilized by Italian clients royalty calculation

Language: Английский

Citations

0