Explorando Modelos com Janelas Temporais para Previsão de Acordes Musicais a Partir de Melodias

• Marlon Duarte UFC
• Raylander Marques UFC
• Gabriel Rudan UFC
• Zairo Bastos UFC
• César Lincoln UFC

DOI: https://doi.org/10.5753/webmedia.2025.16140

Resumo

The task of predicting chords in music is highly important in the careers of composers and professional musicians. Composers create new melodies that require harmony to make them, among many other things, more marketable. Professional arranger musicians frequently need to learn new songs for which they lack supporting material. This work explores chord prediction in melodies using Machine Learning. Utilizing the POP909 dataset 1, composed of 909 songs in MIDI format, Random Forest (RF) and Long Short-Term Memory (LSTM) models were trained and compared, both with and without bidirectionality, employing features such as melody, note intensity, musical key, and the delta time of each melodic interval’s execution. The RF model achieved an overall average accuracy of approximately 77%, performing well on common chords. Conversely, the LSTM without bidirectionality achieved around 61% accuracy, and with the use of the technique (BLSTM), it obtained approximately 73% accuracy. Both models demonstrated challenges in predicting rare chords, such as diminished chords. The contributions aim to advance AI-assisted musical analysis, focusing on applications for composition and live accompaniment.

Palavras-chave: Chord Prediction, Musical Sequences, Machine learning, Time Series

Referências

Carlos V.S. Araujo, Rafael Giusti, and Marco A.P. Cristo. 2018. Um Modelo de Predição para o Sucesso no Mercado Musical. In Anais Estendidos do XXIV Simpósio Brasileiro de Sistemas Multimídia e Web (Salvador). SBC, Porto Alegre, RS, Brasil, 37–42. DOI: 10.5753/webmedia.2018.4559

Adam Patrick Bell. 2018. Dawn of the DAW: The studio as musical instrument. Oxford University Press, Oxford, United Kingdom.

Rolando Benenzon. 1989. Teoria da musicoterapia. Grupo Editorial Summus, São Paulo, SP.

Jean-Pierre Briot and François Pachet. 2020. Deep learning for music generation: challenges and directions. Neural Computing and Applications 32, 4 (2020), 981–993. DOI: 10.1007/s00521-018-3813-6

Kleberson Calanca. 2021. Entre a Era do Rádio e a Bossa-Nova: O avanço das tecnologias de captação, gravação e difusão do som como meios criadores. Revista Brasileira de Música n.d., n.d. (2021), n.d. Informações de volume, número e páginas não disponíveis online.

Fabio Adour da Camara. 2008. Sobre Harmonia: uma proposta de perfil conceitual. Ph.D. Dissertation.

Cedric De Boom, Stephanie Van Laere, Tim Verbelen, and Bart Dhoedt. 2019. Rhythm, chord and melody generation for lead sheets using recurrent neural networks. In Joint European conference on machine learning and knowledge discovery in databases. Springer, Berlin / Heidelberg, Germany, 454–461. DOI: 10.48550/arXiv.2002.10266

Agostino Di Scipio. 2000. The technology of musical experience in the 20th century. Rivista italiana di musicologia 35, 1/2 (2000), 247–275.

Mateus Rocha Grain. 2023. Das memórias que cercam os intervalos de três tons: entre sons e conotações. Ph.D. Dissertation. Universidade do Estado de Santa Catarina - UDESC. Disponível em: [link].

Mark Granroth-Wilding and Mark Steedman. 2014. A robust parser-interpreter for jazz chord sequences. Journal of New Music Research 43, 4 (2014), 355–374. DOI: 10.1080/09298215.2014.910532

Lucy Green, Flávia Motoyama Narita, and Luciana Fernandes Hamond. 2022. O que os professores podem aprender com os músicos populares? Revista Orfeu 7, 1 (2022), 1–14. DOI: 10.5965/2525530407012022e0401

Jack Hardwick. 2024. An LSTM-Based Chord Generation System Using Chroma Histogram Representations. arXiv preprint arXiv:2405.05240 n.d., n.d. (2024), n.d. DOI: 10.48550/arXiv.2405.05240 Informações de volume, número e páginas não disponíveis online.

Cheng-Zhi Anna Huang, Ashish Vaswani, Jakob Uszkoreit, Noam Shazeer, Ian Simon, Curtis Hawthorne, AndrewMDai, MatthewD Hoffman, Monica Dinculescu, and Douglas Eck. 2018. Music transformer. arXiv preprint arXiv:1809.04281 n.d., n.d. (2018), n.d. DOI: 10.48550/arXiv.1809.04281 Informações de volume, número e páginas não disponíveis online.

Yu-Siang Huang, Szu-Yu Chou, and Yi-Hsuan Yang. 2018. Pop music highlighter: Marking the emotion keypoints. arXiv preprint arXiv:1802.10495 n.d., n.d. (2018), n.d. DOI: 10.48550/arXiv.1802.10495 Informações de volume, número e páginas não disponíveis online.

Hyungui Lim, Seungyeon Rhyu, and Kyogu Lee. 2017. Chord generation from symbolic melody using BLSTM networks. arXiv preprint arXiv:1712.01011 n.d., n.d. (2017), n.d. DOI: 10.48550/arXiv.1712.01011 Informações de volume, número e páginas não disponíveis online.

Nolan Monnier, Darien Ghali, and Sophie X. Liu. 2021. FFT and Machine Learning Application on Major Chord Recognition. In 2021 Twelfth International Conference on Ubiquitous and Future Networks (ICUFN). IEEE, Jeju Island, Korea, 426–429. DOI: 10.1109/ICUFN49451.2021.9528762

Amin Nayebi, Sindhu Tipirneni, Chandan K Reddy, Brandon Foreman, and Vignesh Subbian. 2023. WindowSHAP: An efficient framework for explaining time-series classifiers based on Shapley values. Journal of biomedical informatics 144 (2023), 104438. DOI: 10.1016/j.jbi.2023.104438

Pariwat Ongsulee. 2017. Artificial intelligence, machine learning and deep learning. In 2017 15th International Conference on ICT and Knowledge Engineering (ICT&KE). IEEE, Bangkok · Thailand, 1–6. DOI: 10.1109/ICTKE.2017.8259629

Marcelo Brosowicz de Paulo et al. 2022. Geração de Música com Machine Learning. Florianópolis, SC n.d., n.d. (2022), n.d. Trabalho de Conclusão de Curso – Universidade Federal de Santa Catarina (UFSC). Disponível em: [link].

Rodolfo Miranda Pereira and Carlos N. Silla. 2017. Using simplified chords sequences to classify songs genres. In 2017 IEEE International Conference on Multimedia and Expo (ICME). IEEE, Hong Kong, China, 1446–1451. DOI: 10.1109/ICME.2017.8019531

Felix Reuß, Isabella Greimeister-Pfeil, Mariette Vreugdenhil, and Wolfgang Wagner. 2021. Comparison of long short-term memory networks and random forest for sentinel-1 time series based large scale crop classification. Remote Sensing 13, 24 (2021), 5000. DOI: 10.3390/rs13245000

Curtis Roads. 1996. The computer music tutorial. MIT press, Massachusetts, EUA.

Adam Roberts, Jesse Engel, Colin Raffel, Curtis Hawthorne, and Douglas Eck. 2018. A Hierarchical Latent Vector Model for Learning Long-Term Structure in Music. In Proceedings of the 35th International Conference on Machine Learning (Proceedings of Machine Learning Research, Vol. 80), Jennifer Dy and Andreas Krause (Eds.). PMLR, Cambridge MA, 4364–4373. [link]

Nipun Sharma, Swati Sharma, C Bhanuprakash Reddy, Manisha A, Bahuguna P, and Ankita T. 2025. Analysis of Isotonic Calibration on Gaussian Naïve Bayes Performance for Guitar Chords Classification. In 2025 International Conference on Automation and Computation (AUTOCOM). IEEE, Dehradun, India, 15–19. DOI: 10.1109/AUTOCOM64127.2025.10956982

Jáderson Aguiar Teixeira. 2015. O ensino musical interdisciplinar de harmonia, contraponto, solfejo e arranjo como estratégia de produção de conhecimento. www. teses. ufc. br n.d., n.d. (2015), n.d. Informações de volume, número e páginas não disponíveis online. Disponível em [link].

Ziyu Wang, Ke Chen, Junyan Jiang, Yiyi Zhang, Maoran Xu, Shuqi Dai, Xianbin Gu, and Gus Xia. 2020. Pop909: A pop-song dataset for music arrangement generation. arXiv preprint arXiv:2008.07142 n.d., n.d. (2020), n.d. DOI: 10.48550/arXiv.2008.07142 Informações de volume, número e páginas não disponíveis online.

Bruna D Wundervald and Walmes M Zeviani. 2019. Machine learning and chord based feature engineering for genre prediction in popular Brazilian music. arXiv preprint arXiv:1902.03283 n.d., n.d. (2019), n.d. DOI: 10.48550/arXiv.1902.03283 Informações de volume, número e páginas não disponíveis online.