Modeling Real-Time Interactive Conversations as Timed Diarized Transcripts

TL;DR

This paper introduces a method to simulate real-time interactive conversations with pretrained language models by modeling timed diarized transcripts, enabling more natural and timely chatbot interactions.

Contribution

The paper presents a novel approach using timed diarized transcripts and causal rejection sampling to enable real-time conversational modeling with pretrained language models.

Findings

Effective at 30 tok/s for instant messaging

Achieves 20 tok/s for spoken conversations

Runs on commodity hardware with minimal data

Abstract

Chatbots built upon language models have exploded in popularity, but they have largely been limited to synchronous, turn-by-turn dialogues. In this paper we present a simple yet general method to simulate real-time interactive conversations using pretrained text-only language models, by modeling timed diarized transcripts and decoding them with causal rejection sampling. We demonstrate the promise of this method with two case studies: instant messenger dialogues and spoken conversations, which require generation at about 30 tok/s and 20 tok/s respectively to maintain real-time interactivity. These capabilities can be added into language models using relatively little data and run on commodity hardware.