Tencent Video Cloud Editing System: Architecture and Optimization Techniques

This article introduces Tencent Video's cloud editing system designed for online live stream editing. The system addresses the traditional VOD transcoding workflow limitations, where satellite signals are captured, edited offline with non-linear editing software, transcoded, and then uploaded - a process that causes significant delays in program release.

Cloud Editing Concept: The system processes live streams by receiving two parallel streams from the encoder (low-bitrate for online editing, high-bitrate for distributed transcoding) with identical PTS timestamps. This enables real-time editing while the program is live, completing the editing process immediately after the broadcast ends.

Optimization Methods:

1. Fuzzy Seek: When users drag the timeline cursor, only keyframe data needs to be transmitted and decoded, significantly improving response speed. The function OutFrame_key = F(pos) retrieves the nearest keyframe to the user's current position.

2. Frame-Precise Cutting: Online editing uses HLS technology where each segment is a TS file with a 33-bit timecode field (90kHz unit, resetting every 26 hours). To ensure continuous, unique, monotonically increasing timestamps, the system calculates PTS using formulas based on segment duration (#EXT-X-TARGETDURATION) and frame numbers within each segment.

3. WYSIWYG Features: Using NPAPI plugin technology with double-buffering rendering to provide real-time feedback for operations like ad removal, logo insertion, and black border cropping.

Future Direction: Evolving from a specialized live streaming editing tool to a general-purpose cloud editing platform with features like subtitles, special effects, online material libraries, and audio/image synchronization capabilities.