Self-Supervised Syllable Discovery Based on Speaker-Disentangled HuBERT