Scriptora — Context-Aware Realtime Subtitles

Scriptora is a context-aware realtime subtitle system built on AssemblyAI's Realtime Speech-to-Text API. It turns spoken audio into live subtitles and lets users correct transcription errors by voice while the conversation continues. Scriptora combines realtime speech recognition, project-specific vocabulary, and two correction paths. Explicit replacements are handled deterministically for speed and predictability. When a correction requires inference or a naturally worded repair needs to be interpreted, Scriptora uses Qwen through the AssemblyAI LLM Gateway. Project vocabulary can be added during a live session and pushed into AssemblyAI's keyterms_prompt, giving the realtime recognizer additional context for terminology used in that session. When a correction is applied, Scriptora preserves the original transcription and places the corrected version underneath it, making the change transparent rather than silently overwriting what was heard. The system is designed for live situations such as presentations, live captioning, technical discussions, and other workflows where stopping to manually edit subtitles interrupts the flow.