Wire arc additive manufacturing (WAAM) offers advantages such as rapid fabrication and compatibility with lightweight structural design, and thus exhibits great potential for manufacturing critical aerospace components. To address the various defects that may arise during the WAAM process, an unsupervised cross-modal anomaly detection method is introduced, integrating an improved residual deep temporal convolutional network (IRD-TCN) with a residualenhanced lightweight attention (RELA) module for the fusion of acoustic and electrical signals. Owing to the complexity of WAAM operating conditions, the detection capability of a single sensing modality is limited. Therefore, wavelet-based features from acoustic signals are fused with electrical signal features through cross-modal relational analysis, enabling IRD-TCN and RELA to identify changes in the correlation between the two sensing signals and achieve high-quality detection. Experimental results indicate that the proposed approach attains a precision of 98.37%, a recall of 97.73%, and an F1-score of 98.10%, effectively addressing the limitations of traditional data-fusion methods in terms of accuracy and robustness for anomaly identification in WAAM processes.