Distinguishability-Driven Voice Generation for Speaker Anonymization via Random Projection and GMM | Synapse