Signal-to-Noise Ratio Tolerance in Self-Supervised Speech Representations Across Low-Resource Languages