Voice dialog system based on RWKV model

This paper aims to develop an intelligent voice interaction dialogue system for the aging Chinese society, focusing on meeting the needs of the elderly in home-based care. The system is designed with speech recognition, language model, and speech synthesis, using a large-scale language model called RWKV. To address the disadvantage that existing dialogue systems cannot make effective responses for the elderly, this paper is based on the corpus of the elderly and combines the RWKV model, which is fine-tuned by Low-Rank Adaptation (LoRA) technology to generate natural language responses. The experimental results show that the system partially improves the fluency and reasonableness of the answers when dealing with the daily problems of the elderly, and that the intelligent dialogue system based on the RWKV model has a great potential to be applied in the elder's home-based community care, which can improve the accuracy and naturalness of the answers, and will reduce the operation cost to a certain extent. In addition to this, this paper provides a comparison of loss function curves for some of the initial and final learning rates. Finally, future work will further optimise the model to improve its adaptability and fluency in different application scenarios.