kapynResearch

Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers

Apple details the memory-efficient audio architecture powering on-device expressive voices in Siri. The system uses a specialized detokenizer to translate semantic audio tokens from the AFM 3 Core model into high-fidelity speech within strict hardware limits. This research enables sophisticated generative audio capabilities to run efficiently on resource-constrained edge hardware like the Apple Matrix Coprocessor.

Apple ML Research·Jul 28, 2026

Opening Kapyn…