Robust algorithms and interaction strategies for voice spelling

This paper describes our experiments to overcome the problem of inputting unknown words in a voice-only application by spelling. As the paper demonstrates, this is achieved by postprocessing the speech recognition result to attempt to recover from possible recognition errors that occurred during the spelling phase. Since the post-processing does not involve any additional user input, it not only increases the robustness of voice spelling, but it also reduces the cognitive burden put on users and increases the overall usability of systems that require spelling by voice.