pyryt posted https://arxiv.org/abs/2010.10874 https://arxiv.org/abs/2010.10874, which might be helpful here, but we probably end off with personalized models th

pyryt posted https://arxiv.org/abs/2010.10874 https://arxiv.org/abs/2010.10874, which might be helpful here, but we probably end off with personalized models that learned from conversation styles. A magic stop/processing word would be the easiest to add since you already have the transcript, but it's taking the natural feel of a conversation.