Real-Time Transcription in Arbitration Hearings

How TERES Produces a Real-Time Transcript

TERES runs transcription as a managed service rather than as software the parties operate themselves. AI speech-to-text engines generate the transcript live during the hearing, and trained human transcribers review it in real time.

The transcript is verbatim as standard, with speaker identification and timestamps. A cleaned-up version is available on request.

Accuracy: AI Output and Human Review

The TERES service level agreement sets two accuracy figures. AI transcription is at least 98% accurate for clear audio, and transcripts reviewed by TERES human transcribers are 99.5% accurate.

The human review layer is also what keeps speaker attribution and context right when several advocates, witnesses and tribunal members speak in turn.

Searching the Transcript During the Hearing

Because the record exists as the hearing runs, counsel and the tribunal can search and query it while the hearing is still in progress, for example to find what a witness said earlier in cross-examination.

TERES also provides daily summaries of the hearing record and action points drawn from each hearing day.

From Live Feed to Final Transcript

There are two timelines to keep apart. The real-time transcript is available during the hearing itself, and the final version follows within a few hours of the hearing closing.

Where a transcript is instead produced after the hearing rather than live, TERES typically delivers it within 24 to 48 hours, depending on length and complexity. Expedited delivery is available on request, subject to availability.

Confidentiality of the Hearing Record

How the hearing record is stored and protected matters as much as how it is produced. TERES is certified to ISO/IEC 27001:2022 for information security management.

TERES does not share, sell or use client data for any purpose other than delivering the agreed service, and does not use client data to train AI models.

AI Transcription in Institutional Arbitration

At least one arbitral institution has published its own model for AI-assisted transcription. In September 2023 the American Arbitration Association and International Centre for Dispute Resolution (AAA-ICDR) announced an AI-powered transcription service, provided with Optima Juris, in which voice recognition creates the transcript and two levels of human review and editing follow.

Under that published model, rough drafts are available within 24 hours after the proceedings and final transcripts are delivered within three business days, and the AAA-ICDR states that AI-powered transcriptions are permissible under its rules. The announcement does not describe a live transcript feed. It is cited here as context on institutional practice; it does not refer to TERES.

Languages

TERES transcription is English-primary. Other languages can be supported on request, depending on the availability of trained resources.

Questions to Settle Before the Hearing

When agreeing transcript arrangements with the tribunal and the other side, it helps to confirm:

• Whether the tribunal and counsel need a live feed during the hearing, or only a transcript afterwards
• Which accuracy figure applies: AI output alone, or the transcript after human review
• When the final version of each day is due, and whether expedited delivery is needed
• Whether a verbatim or a cleaned-up transcript is required
• How the record is stored and protected, and whether the provider uses client data to train AI models