Formula-governed model/device/system-agnostic voice coprocessor architecture using capability-negotiated universal voice events, replaceable acoustic engines, portable interruption/control, AEC/double-talk handling, platform adapters and evidence.