It sounds like you’ve already tried several approaches, so I’d start by checking the segmentation settings and the limitations of the algorithm before retraining DCL. The 8,192-sample limit could be relevant if your audio events are longer than the supported segment size, but it would be worth confi...