Voiceprint Division Multiplexing for Parallel Speech Transmission
Zhen Qiang Gao, Lini Yuan, Xiaodong Xu, Zhu Han · 2023
Fast development of mobile communication makes spectrum resources getting increasingly scarce, and the growing number of user accesses can lead to network congestion. On the other hand, speech separation is a classic research topic in the field of natural language processing. With the fast development of deep learning, the performance of speech separation technology has been continuously improved. In this study, we introduce the speech separation technology into the mobile communication system to improve spectrum efficiency. In particular, Voiceprint Division Multiplexing (VDM) is proposed to utilize the same physical channel for parallel transmission of multiple speech signals. We first introduce the system structure and working procedure of VDM, and then established a simulation platform with the classic speech separation model and multi-bitrate voice codec to validate the effectiveness of the proposed scheme. Experimental results on the LibriSpeech dataset indicate that the proposed parallel speech transmission method outperforms the normal separate transmission scheme in terms of Scale-Invariant Source-To-Noise Ratio (SI-SNR) when the bitrate is below 9 kbps, and the maximum improvement can be 4.4 dB. This improvement can be further improved to 6.6 dB for low bitrate by smart grouping of multiple speech signals, which demonstrates the rationality and feasibility of VDM.