Speaker embeddings by modeling channel-wise correlations

Re-assign community