Nanopore sequencing enables real-time nucleic acid analysis. Its primary output is a one-dimensional electrical signal, traditionally converted into nucleotide sequences with basecalling before further analysis. However, basecalling is a computational bottleneck, prompting the development of methods that segment the signal into events corresponding to individual nucleotide translocations and process them directly. Traditional algorithmic segmentation approaches often struggle with robustness, failing to capture the diverse characteristics of nanopore signals and introducing errors in noisy regions. To overcome these challenges, we introduce Campolina, a novel deep-learning framework that directly processes raw nanopore signals to produce accurate segmentation. We define a set of metrics to ensure a comprehensive and fair assessment of segmentation quality and demonstrate that Campolina significantly enhances segmentation quality and improves the performance of existing signal-processing frameworks across various tasks.