| description abstract | Voice is the most convenient means for human beings to communicate with others, even if the objects of their communication are not other humans but machines or computers. Many industries, and even the architecture, engineering, construction, and operations (AECO) industry, have attempted to study and apply speech recognition systems in their operations to improve work efficiency and productivity. However, previous studies on speech recognition had two limitations: they used keywords requiring basic knowledge of building information modeling (BIM) commands for using them and in searching BIM data, they relied on the Industry Foundation Classes (IFC) format, which involves converting BIM data to IFC. Such methods did not conduce to direct retrieval in BIM software. In the latter case, data search was possible, but data manipulation was not. To improve on the limitations of previous studies, this study developed a building information modeling automatic speech recognition (BIMASR) framework that requires no knowledge of BIM commands, which allows for the input of natural language (NL)-based questions into BIM software using human voice to search and manipulate data. The framework consists of three modules: one for voice recognition, one for natural language processing (syntax and semantic analysis), and one for BIM data preprocessing and interworking with relational databases. The manipulation of BIM data with NL-based speech recognition converts the BIM operating environment from an expert-oriented into a user-oriented environment. This conversion allows for more BIM interaction and the popularization of BIM use and enhances the use of BIM in dynamic environments such as virtual reality, augmented reality, and holograms, where conventional input devices are typically absent. | |
