Abstract: 【Purpose/significance】Integrate the textual and phonetic features of ancient books to analyze and study them in a multi⁃ modal manner【. Method/process】Firstly, BERT was used to extract text features, and MFA voice forced alignment model and Librosa audio processing tool were used to extract audio features. Then, text features and audio features are fused in the multimodal fusion layer. Finally, the fusion features are input into BiLSTM-CRF layer for label prediction and output the results, and the ancient texts analysis model with audio features(TAMAF) is constructed.【Result/conclusion】After incorporating appropriate audio features, the pro⁃ posed model outperformed the baseline model in all four downstream validation tasks. Among them, the highest improvement in sen⁃ tence breaking effect was 8.54%; The highest improvement in word segmentation effect was 0.21%; The highest improvement in named entity recognition was 0.97%; The highest improvement in part of speech tagging performance was 0.85%. Therefore, the TAMAF model proposed in this paper is superior to past models, which can capture the interaction between modalities and improve the process⁃ ing effect of ancient books【. Innovation/limitation】In the field of speech processing, audio features that express other physical mean⁃ ings can be integrated into models for exploration and analysis. In addition, audio and text features can be better integrated and inter⁃ acted on a wider dataset.