Watermark Synchronization for Feature-Based Embedding: Application to Speech