Estonian courts are using AI transcription services

Summary:

Using AI-driven solutions, courts can automate the transcription process, significantly reducing the time needed to produce accurate records. 

Client: 

Estonian Courts

Problem Statement: 

Traditional manual transcription during court hearings is slow, labor-intensive, and prone to human error, making it difficult to keep up with the high volume of court cases. 

While speech recognition technology has been around for years, it is only with recent advancements in ML that it has become feasible to produce accurate, automated transcripts. This innovation greatly streamlines the preparation of court documents by reducing the time and effort required. In response to these developments, the Ministry of Justice of Estonian launched the creation of AI solution, a custom-built speech recognition solution tailored for court use. 

 

Results:

  • Enhanced efficiency and accuracy in transcribing Estonian court hearings.
  • High-quality transcripts generated within seconds.
  • Optimized use of time and resources.
  • Simplified administrative processes.
  • Freed up court staff to concentrate on more critical tasks.

AI Solution Overview:

Tilde collaborated with CGI Estonia to create Salme, a custom automatic speech recognition (ASR) solution designed specifically for Estonian court hearings. Salme enables real-time and offline transcription of court proceedings at all levels by converting speech into text.

Tilde focused on developing the ASR system, which handles transcription in both real-time and offline modes. CGI contributed by designing the Windows Presentation Foundation (WPF) user interface and integrating the audio recording system with Tilde’s ASR technology. Additionally, CGI managed the integration of Salme with Estonia’s Court Information System through X-Road®—an open-source platform that ensures secure and unified data exchange between organizations. This integration facilitated the seamless and automated transfer of information.

The speech recognition model was fine-tuned using more than 800 hours of transcribed audio and a textual dataset of over 800 million words, provided by the client. Since its launch, Salme has been successfully deployed across all national and regional courts in Estonia.

References: 

  1. Estonian Courts shift to automated transcription with Salme. https://tilde.ai/asr-solution-for-estonian-courts/

Industry: Legal Services

Vendor: CGI Estonia and Tilde

Clients: Estonian Courts

Publication Date: 2024