Singing voice timbre anomaly detection based on attention mechanism and STFT time-frequency graph processing

With the advancement of big data intelligent algorithms, the detection accuracy of bel canto vocal timbre has gradually improved, yet challenges persist in low precision and heavy reliance on sample data. To address these limitations, this study proposes a detection model integrating Fourier transform time-frequency diagrams with a hybrid attention mechanism, incorporating meta-learning algorithms to reduce sample dependency. Comparative results against three existing models on self-collected datasets demonstrate that our model achieves 97.23% classification accuracy, with error rates and minimum joint detection cost index reaching 4.23% and 2.56% respectively. The Euclidean distance between normal and abnormal timbres reaches 20.298, showcasing superior anomaly detection capabilities. These findings indicate that the proposed model significantly enhances the accuracy of bel canto vocal timbre detection, providing valuable technical references for bel canto vocal training.