Timeout on transcriptions using speaker labels

Updates timeline

1. Investigating Jun 25, 03:37 PM UTC

   We are currently investigating an issue with async transcriptions on the US endpoint that use speaker labels. This is resulting in timeouts for all transcriptions with speaker labels enabled for files under 2 minutes long. This began at 15:00 UTC.

2. Identified Jun 25, 03:38 PM UTC

   The issue has been identified and a fix has been released.

3. Monitoring Jun 25, 04:00 PM UTC

   A fix has been implemented and our API is slowly beginning to recover normal processing.

4. Monitoring Jun 25, 04:14 PM UTC

   The root cause has been addressed and our system is slowly working through the transcript backlog, which is causing slower processing on our US async endpoint.

5. Monitoring Jun 25, 04:27 PM UTC

   Our API is continuing to recover and incoming transcriptions are returning to normal processing. We are continuing to monitor the situation.

6. Resolved Jun 25, 04:48 PM UTC

   This incident has been resolved.

7. Postmortem Jun 25, 06:02 PM UTC

   **What happened** An issue related to our speaker diarization service caused certain requests to hang until they timed out. This lead to a backlog of requests to our async API creating a processing queue across our US transcription pipeline, which slowed and timed out transcription requests more broadly, including requests that were not using speaker diarization. --- **When** Impact ran from approximately 15:00 to 16:25 UTC on June 25, 2026 \(about 1 hour 25 minutes\). The most severe disruption was in the first 45 minutes \(15:00–15:45 UTC\); remaining delays and timeouts cleared shortly after. Full recovery was confirmed by ≈16:40 UTC. --- **What services were affected** On our US async API endpoint transcription requests during this window saw processing delays and timeouts. Our EU async API endpoint was not affected and operated normally throughout.