A Comparative Analysis of Machine Learning Models for Acoustic Classification: A Reproducible Modular Pipeline Across Three Domains

• Filipe Gomes Ferreira FATEC
• Denival Biotto Filho FATEC / IFSP
• Luan Vitor Pereira Rocha FATEC

DOI: https://doi.org/10.5753/eniac.2026.30694

Resumo

A classificação acústica aparece em domínios distintos, da manutenção preditiva ao monitoramento bioacústico e à análise de sons urbanos, mas a maioria dos estudos avalia um único modelo sobre uma única base. Este artigo apresenta um pipeline modular e reprodutível que extrai MFCCs, espectrogramas Mel e descritores espectrais e temporais, e o usa para comparar seis modelos (Árvores de Decisão, Random Forests, SVM, MLP, CNN e aprendizado por transferência com YAMNet) sob um mesmo protocolo em três domínios com dados públicos: aves (Xeno-canto), sons urbanos (UrbanSound8K) e válvulas industriais (MIMII). A CNN sobre espectrogramas Mel liderou em todos os domínios (até 99,6%), os modelos clássicos enriquecidos ficaram a poucos pontos a um custo bem menor, e o aprendizado por transferência não foi uniformemente superior, o que discutimos junto a uma assimetria de erro relevante para detecção de falhas.

Referências

Choi, K., Fazekas, G., Sandler, M. and Cho, K. (2017) "Convolutional Recurrent Neural Networks for Music Classification", In: IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 2392-2396.

Eyben, F., Scherer, K. R., Schuller, B. W., et al. (2016) "The Geneva Minimalistic Acoustic Parameter Set (GeMAPS) for voice research and affective computing", IEEE Transactions on Affective Computing, v. 7, n. 2, p. 190-202.

Ferreira, L. L., Oliveira, A., Teixeira, N., Bulut, B., Landeck, J., Morgado, N. and Sousa, O. (2021) "Predictive Maintenance of home appliances: Focus on Washing Machines", In: Proceedings of the School of Engineering of the Polytechnical Institute of Porto / University of Coimbra.

Gong, Y., Chung, Y.-A. and Glass, J. (2021) "AST: Audio Spectrogram Transformer", In: Proceedings of Interspeech.

Hershey, S., Chaudhuri, S., Ellis, D. P. W., et al. (2017) "CNN Architectures for LargeScale Audio Classification", In: IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 131-135.

Logan, B. (2000) "Mel Frequency Cepstral Coefficients for Music Modeling", In: International Symposium on Music Information Retrieval.

McFee, B., Raffel, C., Liang, D., Ellis, D. P. W., et al. (2015) "librosa: Audio and music signal analysis in Python", In: Proceedings of the 14th Python in Science Conference, p. 18-25.

Mobley, R. K. (2002) "An Introduction to Predictive Maintenance", 2nd edition, Butterworth-Heinemann.

Purohit, H., Tanabe, R., Ichige, K., Endo, T., Nikaido, Y., Suefusa, K. and Kawaguchi, Y. (2019) "MIMII Dataset: Sound dataset for malfunctioning industrial machine investigation and inspection", In: Proceedings of the 4th Workshop on Detection and Classification of Acoustic Scenes and Events (DCASE).

Salamon, J. and Bello, J. P. (2017) "Deep Convolutional Neural Networks and Data Augmentation for Environmental Sound Classification", IEEE Signal Processing Letters, v. 24, n. 3, p. 279-283.

Salamon, J., Jacoby, C. and Bello, J. P. (2014) "A Dataset and Taxonomy for Urban Sound Research", In: Proceedings of the 22nd ACM International Conference on Multimedia, p. 1041-1044.

Stowell, D. (2022) "Computational bioacoustics with deep learning: a review and roadmap", PeerJ, v. 10, e13152.

Tsalera, E., Papadakis, A. and Samarakou, M. (2021) "Comparison of pre-trained CNNs for audio classification using transfer learning", Journal of Sensor and Actuator Networks, v. 10, n. 4, p. 72.

Xeno-canto Foundation. (2026) "Xeno-canto: Sharing bird sounds from around the world", [link], Janeiro.