【學術亮點】Deep learning–empowered triboelectric acoustic textile for voice perception and intuitive generative AI-voice access on clothing
Facility Agricultural: Green Energy Development and Carbon Offset【Department of Materials Science Engineering / Lai, Ying-Chih / Distinguished Professor】
設施農業：農業綠能開發與碳匯補償【材料科學與工程學系賴盈至教授/特聘教授】
| 論文篇名 | 英文：Deep learning–empowered triboelectric acoustic textile for voice perception and intuitive generative AI-voice access on clothing 中文：深度學習賦能的摩擦電聲學紡織品，用於語音感知和服裝上直觀的生成式人工智慧語音訪問 |
| 期刊名稱 | Science Advances |
| 發表年份,卷數,起迄頁數 | 22025, 11(41), no. eadx3348 |
| 作者 | Shao, Beibei; Wu, Tai-Chen; Yan, Zhi-Xian; Ko, Tien-Yu; Peng, Wei-Chen; Jhan, Dun-Jie; Chang, Yu-Hsiang; Fong, Jiun-Wei; Lu, Ming-Han; Yang, Wei-Chun; Chen, Jiann-Yeu(陳建宇); Lu, Ming-Yen; Sun, Baoquan*; Liu, Heng-Jui*; Liu, Ruiyuan*; Lai, Ying-Chih(賴盈至)* |
| DOI | 10.1126/sciadv.adx3348 |
| 中文摘要 | 將生成式人工智慧聊天機器人與聲學感知紡織品結合，使日常服裝能夠透過語音互動檢索資訊、尋求建議並執行任務。在此，我們展示了首款基於深度學習 (DL) 的摩擦電人工智慧聲學紡織品 (A-Textile)，它利用服裝上的靜電荷實現潛移默化的主動語音感知和人工智慧存取。多層 A-Textile 採用嵌入矽橡膠的三維 SnS2 奈米花 (NFs) 複合塗層 (SR；SnS2 NFs-SR)，以增強電荷的捕獲和傳輸，同時也採用了 SnS2 NFs 修飾的類石墨碳化紡織品 (SnS2 NFs-GT)，用於電荷的積累和保存。該設計最大限度地提高了紡織品上的電荷密度，實現了 21 V 的輸出、1.2 V Pa−1 的靈敏度、1 Hz 的分辨率以及 80–900 Hz 的寬聲響應頻率範圍。 A-Textile 使用訓練有素的深度學習模型，對語音命令進行精確分類和視覺化，以實現物聯網控制和雲端資訊存取。此外，我們還示範了它與 ChatGPT 的集成，從而提供了一個直覺的介面，方便使用者使用生成式 AI 服務執行複雜的任務。 |
| 英文摘要 | Integrating generative AI chatbots with acoustic perception textiles allows everyday clothing to retrieve information, seek advice, and perform tasks through voice interactions. Here, we present the first deep learning (DL)-empowered triboelectric AI acoustic textile (A-Textile) leveraging electrostatic charges on clothing for imperceptible, active voice perception and AI access. The multilayered A-Textile features a composite coating of three-dimensional SnS2 nanoflowers (NFs) embedded in silicone rubber (SR; SnS2 NFs-SR) to enhance charges capture and transfer, along with a SnS2 NFs-decorated graphite-like carbonized textile (SnS2 NFs-GT) for charge accumulation and preservation. This design maximizes the charge density on the textile, achieving 21 V output, 1.2 V Pa−1 sensitivity, 1-Hz resolution, and a wide sound response frequency range of 80–900 Hz. Using a well-trained DL model, the A-Textile precisely classifies and visualizes voice commands for Internet-of-Things control and cloud information access. Furthermore, we demonstrate its integration with ChatGPT, providing an intuitive interface for engaging with generative AI services to perform sophisticated tasks. |
| 發表成果與本中心研究主題相關性 | 新能源採集 |
