Semi-Automated Relay Captioning with ASR-Human Switching

The system merges automated transcription software with human operator assistance in a hybrid architecture. The automated software handles initial transcription to maintain speed, while human operators review and correct errors to ensure accuracy. This combination allows the system to achieve both high transcription speed and high accuracy simultaneously.

Human operators perform only partial transcription work rather than complete transcription. The system uses automated software for the majority of transcription tasks, with human operators intervening only when needed to correct errors or handle difficult cases. This partial human involvement maintains accuracy while significantly reducing operational costs compared to fully manual transcription.

The system performs speaker training in advance before transcription is needed. Voice prints and speaker characteristics are captured and stored in a database during off-peak times or initial setup phases. When transcription is required, the pre-trained models are immediately applied, eliminating training delays during actual transcription operations.

The system introduces a intelligent router as an intermediary component that automatically directs transcription tasks to either automated software or human operators based on task characteristics, confidence levels, and resource availability. This intermediary manages the complexity of coordinating multiple transcription methods, handling quality assurance, and optimizing resource allocation, thereby reducing operational costs while keeping the system architecture manageable.