Aims: Signal quality remains a critical challenge in ambulatory ECG (Holter) recordings. Artifacts may significantly downgrade the performance of rule-based and deep learning ECG algorithms, making noise detection a prerequisite for reliable analysis. We introduce a new annotated Holter ECG database designed for signal quality assessment, along with preliminary deep learning results. Methods: The database comprises 319 two-lead Holter ECG recordings from 258 subjects, sampled at 200 Hz. On each Holter lead signal quality was sorted by expert into three classes: Clean, N1 (noisy but still with identifiable QRS complexes), and N2 (major, uninterpretable noise ). The dataset was divided into a training set comprising 200 full recordings and a test set of 1,850 five-second segments extracted from the remaining 119 recordings. The sampling strategy overrepresented complex cases—such as severe noise with ectopic beats (e.g., premature ventricular and supraventricular contractions)—to support robust model training. As a first benchmark on this database, a 1D-ResNet architecture was trained and evaluated for automated signal quality classification. Results : The model achieved a three-class accuracy of 84.7%, with F1-scores of 91.4% for Clean, 79.4% for N1, and 55.7% for N2 segments. Clean segment detection was supported by a sensitivity of 98.2%. In binary classification (Clean vs. N1 or N2), accuracy improved to 90.0% with a predictive value of 97.4% and an F1-score of 88.0% for noise detection. Conclusion: We introduce a new Holter ECG database for signal quality assessment and report benchmark results using a deep learning approach. The deep-learning model achieves 90.0% binary accuracy while preserving 98.2% of clean signals, demonstrating the feasibility of automated artifact screening for clinical Holter monitoring. Differentiating noise severity (N1 vs. N2) remains an open challenge and an area for future investigation. The database will be made available at PhysioNet to support reproducible research in ECG quality assessment.