I&T Wish - Transcription Tool for the Video-Recording Interview System 2026-11-3
Transcription Tool for the Video-Recording Interview System
I&T Wish
Transcription Tool for the Video-Recording Interview System (REF: W-0624)
Matched I&T Solution
Trial Project
Summary and Challenges
This project aims to deploy an on-premises transcription tool within Immigration Department (ImmD) premises to support Video-Recording Interview (VRI) systems. The primary challenge is processing a high annual volume of 10,000 recording hours while strictly maintaining data security through an offline, internally hosted infrastructure. To address this, the solution directly converts MP4 video files from DVDs into formatted Word documents. A key technical highlight is its mixed-language (code-switching) transcription capability, enabling seamless recognition of continuous speech containing primary languages (Cantonese and English) alongside secondary languages (Indonesian or Filipino) within a single clip without manual switching or re-segmentation. The system guarantees an accuracy rate of 85% or above—measured by Character Error Rate (CER) or Word Error Rate (WER)—across six target languages: Cantonese, Mandarin, English, Indonesian, Filipino, and Vietnamese. Additional features include automatic punctuation insertion and customizable dictionaries for domain-specific hotwords.
Expected Outcome
The system significantly enhances operational efficiency by reducing the manual effort and time required by officers to transcribe lengthy recording sessions into written statements, thereby streamlining case processing. To ensure high-fidelity documentation suitable for official record-keeping, it achieves a minimum 85% accuracy rate—measured by Character Error Rate (CER) or Word Error Rate (WER)—across all six target languages: Cantonese, Mandarin, English, Indonesian, Filipino, and Vietnamese. A key capability is seamless multi-lingual speech processing, which accurately captures code-switching and dynamic language mixing (e.g., Cantonese, English, and secondary languages like Indonesian or Filipino) within a single video clip at 85% accuracy without requiring manual intervention or segment splitting. Operating fully offline on local servers within ImmD premises, the deployment strictly complies with government information security and privacy policies. Finally, it provides standardized, ready-to-use Word outputs complete with automatic punctuation and domain-specific terminology configured via customized dictionaries.
Expected Trial Duration
1-month
Contact Information
I&T Wish Proposer
:
Electrical and Mechanical Services Department (EMSD)
