The rise of wearable smart devices raises unprecedented opportunities for self-improvement through ubiquitous behavior tracking and guidance. However, the design of effective wearable behavior intervention systems remains relatively unexplored. To address this gap, we conducted controlled studies focusing on the reduction of unwanted words (e.g., filler words, swear words) in daily communication through auditory feedback using wearable technology. We started with a design space exploration, considering various factors such as the type, duration, and timing of the auditory feedback. Then, we conducted pilot studies to reduce the space of design choices and prototyped a system called WSCoach (Wearable Speech Coach), which informs users when they utter unwanted words in near-real-time. To evaluate WSCoach, we compared it with a state-of-the-art mobile application supporting post-hoc conversation analysis. Both approaches were effective in reducing the occurrence of unwanted words, but WSCoach appears to be more effective in the long run. Finally, we discuss guidelines for the design of wearable audio-based behavior monitoring and intervention systems and highlight the potential of wearable technology for facilitating behavior correction and improvement. For supplementary material, please see the META Appendix and our OSF project at https://osf.io/6vhwn/?view_only=489498d3ac2d4703a17475fc6ca65dfa.