It's a familiar problem. A recording begins on Zoom, Teams, etc, long before the actual presentation starts. Lots of dead air, twiddling thumbs, and "should we start"s from the moderator. This tool automates the trimming of that content, using audio and transcription analysis to detect actual start of presentation, as well as computer vision to extract key info from slides, providing producers with a properly trimmed and named video file.
Manually trimming hundreds of recordings was time-consuming and error-prone. Additionally, video files were not named in a way that made clear which file was for which session.
Intelligent content detection system using combined visual and audio cues to identify the actual beginning of each presentation, as well as the session number.
After development, delivered properly trimmed recordings to clients within a day, and automated the batch upload to their YouTube channel.
Computer vision pipeline detecting black screens and branding, combined with audio silence detection, and extracting session number from title slide shown in each video. The script then automatically trimmed the video to the actual start time of the presentation, and named the trimmed file with the session number.
For each conference, it saved at least a week of manual cleanup work.