Local Realtime Subtitle Translator

Local Realtime Subtitle Translator is a Windows-first desktop tool that captures system audio, recognises English speech and displays English-Chinese subtitles in a resizable overlay. After the speech and translation models have been downloaded, the complete recognition and translation path can run locally.

Realtime pipeline

• WASAPI loopback capture reads audio from meetings, videos, podcasts and live streams without a virtual sound card.
• Short sliding windows and silence-aware segmentation produce provisional text quickly, then replace it with a stable final sentence.
• Faster-Whisper provides fast, balanced and accurate recognition profiles; MarianMT performs offline English-to-Chinese translation.
• A configurable glossary corrects names and specialist terms after translation.

Designed for continuous use

Audio capture, segmentation, recognition, translation and interface updates run as separate queued stages, so one slow operation does not freeze the entire subtitle path. The overlay supports bilingual or Chinese-only display, adjustable type size and opacity, compact and reader layouts, bounded history, and resource-saving runtime profiles for lower-powered computers.

Environment checks, model download helpers and a synthetic-audio mode make the system easier to install and test even when a real loopback device is unavailable.