Multi-Source Domain Adaptation and Fusion for Speaker Verification