How AI is used

A speech-to-text provider creates timestamped words and anonymous speaker clusters. Scheduled unattended hearings receive a second post-hearing diarization pass that can revise anonymous speaker labels before the brief is generated. Transcripts, punctuation, names, and speaker boundaries can still be wrong, particularly with poor audio, cross-talk, accents, and specialized terms. Temporary source audio used for that pass is handled as described in the Privacy Policy.

Configured watch terms and aliases are matched locally with normalization for case, punctuation, possessives, and phrases. A match shows that the configured language appeared in the finalized machine transcript; it does not establish meaning, endorsement, or materiality.

A generative model summarizes the processed transcript and must cite passages in the saved record. The application rejects unknown references, but a valid citation does not guarantee that the summary interpreted the passage correctly. Users should verify consequential content against the source recording.