Clustering baby cry sounds can provide valuable insights into infant needs and potential health conditions. However, selecting the optimal clustering method for such an acoustic dataset remains a challenge. This study explores various unsupervised clustering techniques to determine the most effective approach for grouping baby cries based on their acoustic features. We evaluate methods including K-means, hierarchical clustering, DBSCAN, spectral clustering and Self-Organizing Maps (SOM), analyzing their performance in terms of cluster separation and consistency. A key focus is on assessing clustering validity using internal metrics such as the Silhouette Score and Davies-Bouldin Index. Our findings indicate that certain methods, particularly SOM combined with K-means, provide well-separated clusters, though challenges remain in ensuring robustness across different cry patterns. The results contribute to the broader understanding of infant vocalization analysis and offer a foundation for future studies in automated baby cry classification.