Automated Captioning

Captioning greatly facilitates participation in mainstream society for the hearing impaired. Though enabling software can help, captioning has remained a labor¿intensive and time-consuming process. Commercial applications of automated speech recognition have typically focused on dictation and speech-interactive services, but little effort has been directed towards applying speech technology to the automation of captioning to address high costs and lengthy turnaround times. Commercial recognition software still requires either a restricted speaker set or a restricted linguistic domain; these represent significant barriers to automated captioning efforts.We propose to apply automatic speech processing techniques to the automation of captioning, thereby significantly lowering costs and turnaround times. Our approach utilizes a known text transcript of the program audio to avoid performing recognition. Instead, we focus on determining the optimal alignment between the transcript and the program audio. This tightly constrains the problem domain, and avoids high error rates.Our Phase I prototype showed the feasibility of automating captioning using this approach. In Phase II, we will complete our research and development and incorporate enhancements based on trial feedback, including more sophisticated acoustic models robust to real-world data, an improved user interface, performance upgrades based on new user data, and search and indexing capabilities.