An enhanced feature augmentation and concatenation method for environmental sound recognition using machine learning
SIGNAL IMAGE AND VIDEO PROCESSING, cilt.20, sa.9, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 20 Sayı: 9
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s11760-026-05549-2
- Dergi Adı: SIGNAL IMAGE AND VIDEO PROCESSING
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Compendex, INSPEC, zbMATH, Technology Collection (ProQuest)
- Kayseri Üniversitesi Adresli: Evet
Özet
The classification of environmental sounds represents a significant and challenging subject within the domain of sound recognition, exhibiting a more complex structure compared to organized sounds such as speech and music. This inherent complexity poses challenges for recognition and classification processes. To address this problem, it is necessary to increase the diversity of features or apply data augmentation techniques. In this study, the ESC-10 and ESC-50 datasets, which are widely used in environmental sound classification, were utilized. In the initial phase, Gaussian noise addition, pitch shifting, and time stretching techniques were applied for data augmentation due to the limited amount of data. Consequently, each training sample was expanded into six versions (original + 5 augmentations). The extracted features include MFCC, Mel spectrogram, RMS energy, spectral centroid, spectral bandwidth, chroma-based features, and spectral contrast. A series of pre-processing steps was applied to the resulting feature vector, and the performance was evaluated using different machine learning (ML) models in terms of both accuracy and efficiency. The results indicate that the Extra Trees (ET) achieved the best performance, with an accuracy of 92.5% on the ESC-10 dataset. The precision, recall, and F1-score values were obtained as 94.16%, 92.5%, and 92.32%, respectively. In the ESC-50 dataset, an accuracy of 72.0% was achieved. This study demonstrates that an efficient and high-performance model can be developed using data augmentation techniques and machine learning algorithms for environmental sound classification.