DEGSTalk: Decomposed Per-Embedding Gaussian Fields for Hair-Preserving Talking Face Synthesis | IEEE Conference Publication | IEEE Xplore