Low Bitrate Codec with Joint Vector Quantization and Hoffman Coding for Internet of Things
Dingwei Peng, Xuelin Yuan, Mingjun Ouyang, Qizhen Weng, Ting Xie, Ningze Zhong, Xiangwei Zhu · 2024
Tiantong-1 satellite offers voice communication rates of 0.6, 1.2, and 2.4 kbps, playing a crucial role in emergency scenarios like field rescue. In the context of BeiDou 3, the communication rate for short messages is typically below 1 kbps. It is necessary to compress the audio to a bit rate lower than 1 kbps in satellite audio internet of things. For satellite voice transmission, the conventional vocoder MELPE, based on the source-filtering assumption, is commonly used. However, it falls short in delivering high-quality voice. This paper introduces an end-to-end vocoder for satellite voice transmission, employing a single model that generates different bit rates with residual vector quantization technique. To further reduce the coding rate, we propose joint vector quantization that achieves a 50% reduction in bitrate. Experimental results show that this method can improve the audio quality at 0.50kbps while it remains the same RTF score. We also count the probability of tokens for each codebook to reduce the code rate by Hoffman coding. Experimental results show that Hoffman coding can reduce the coding rate by 14% while remain the same RTF score.