Give required voice scripts one speech owner
In a Pipecat flow with continuous model history, both model text and native actions can emit the same required script. If only native emission arms playback tracking, a model-spoken introduction can finish while the flow still marks it owed; identity confirmation then queues it a…