Speech quality is a major indicator of the quality of service that describes the performance of speech communication network. Intrusive speech quality assessment generally requires a clean reference speech for evaluation, which is not available in applications such as online conferencing. Although the subjective measure of Mean opinion score (MOS) is widely used for assessing speech quality, the process of MOS test is time-consuming and expensive. In this paper, we propose a MOS prediction network for non-intrusive speech quality assessment in online conferencing, which consists of acoustic encoder, time-dependency network and prediction network with specific design. Accommodated with the large-scale dataset including MOS annotations from Interspeech ConferencingSpeech 2022 Challenge for supervised training, the proposed model is capable of predicting the MOS score of the degraded speech in the automatic manner, without the need for human judges and clean reference speech. Our results show that the proposed model is competitive over the baseline method of the challenge in all evaluation metrics.