Piano Transcription Algorithm Based on Self Attention Deep Learning Network

Le résumé fourni par la source

Music transcription is an important means of recording and inheriting music culture. However, existing music transcription algorithms still have certain errors in practical applications. In response to this issue, the study adopts constant Q conversion to process music signals, introduces note start and frame level pitch recognition modules and transfer window attention, constructs a temporal harmonic diagram for music melody extraction, and uses saliency function for music melody smoothing. The experimental results show that at a frequency point of 600 and a search range of 0.5, the overall accuracy of the transcription algorithm is 2.58% and 2.35% higher than other algorithms, and the original pitch accuracy is 2.23% and 1.06% higher, respectively. The accuracy, recall, and F1 score of the transcription algorithm are 2.11%, 2.27%, and 2.21% higher than the second best algorithm, respectively. After removing window attention and recognition modules, the accuracy of the algorithm decreases by 8.07% and 16.76%, respectively. From this, it can be concluded that the piano music transcription algorithm can effectively raise the accuracy of music recognition and transcription, quickly and accurately converting relevant audio into corresponding notes.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.