A novel contrastive learning-based approach to extract disentangled style, emotion, and speaker representations from speech, enabling multi-speaker expressive speech synthesis.
