Sarcouncil Journal of Engineering and Computer Sciences

Abstract: The rapid evolution of regulatory requirements necessitates near-immediate access to searchable financial and healthcare communications transcripts. This article presents a novel on-path architecture that leverages SIP Recording (SIPREC) to mirror Real-time Transport Protocol (RTP) streams to edge-resident GPU clusters for automatic speech recognition. The architecture incorporates a lightweight envelope to maintain packet sequence integrity, enabling cryptographic binding between transcripts and original media files for evidentiary purposes. Adaptive load-shedding mechanisms ensure conversational quality remains uncompromised during peak audit periods, while a configurable policy layer enforces regional data-sovereignty requirements and supports selective keyword redaction. Operational safeguards, including GPU admission controls, model version management, and anomaly detection systems, create a comprehensive framework for auditable and cost-aware compliance mechanisms without degrading network performance. The proposed architecture addresses existing challenges in the IETF MEDIACTRL standards and establishes a foundation for next-generation compliance systems that balance regulatory demands with operational efficiency.