Low-Latency Real-Time Audio Game Commentary System via LLM-Based Parallel Text Generation

The article presents a low-latency real-time audio commentary system for video games that utilizes a parallel text generation approach to reduce inter-utterance silence from 9.6 seconds to 0.3 seconds. By generating text concurrently with speech playback and buffering multiple candidate utterances, the system significantly improves the naturalness of commentary, achieving over a 40% enhancement in similarity to professional speaking patterns. This innovation is crucial for developers aiming to create immersive gaming experiences with real-time, responsive audio commentary.