Towards hate speech detection in low-resource languages: Comparing ASR to acoustic word embeddings on Wolof and Swahili