LLM-Based Bitrate Selection Method for CMR-Supported Audio Codec
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Recent advances in AI-driven network algorithms have aimed to enhance user experience under diverse network conditions. However, designing and validating optimal AI models tailored to specific network environments and objectives remains a challenge for developers. To address this, frameworks such as netLLM [1], which integrates pretrained large language models (LLMs) with multimodal encoders, network heads, and low-rank adaptation (LoRA) [2], have been proposed and shown to improve performance through offline fine-tuning. In this paper, we extend the concept of netLLM to the bitrate optimization problem of audio codecs supporting Codec Mode Request (CMR). Unlike conventional offline training approaches, we propose an online learning-based bitrate selection method that dynamically adapts to fluctuating real-time communication environments. Our method leverages pretrained LLMs to adjust audio stream bitrates according to network conditions, achieving significant improvements in speech quality, as measured by the E-MODEL R-factor [4], compared to conventional AI models [7]. This study demonstrates an effective approach to audio streaming optimization through online fine-tuning, offering a novel pathway for improving speech quality under variable network conditions.
Fields of Study & Subjects
Speech and Audio Processing
Network Traffic and Congestion Control
Speech Recognition and Synthesis
Publication details
- Journal / source
- International Conference on Consumer Electronics
- Pages
- 1–5
- Publication date
- 3 February 2026
- Language
- EN