From 21:51 PT on November 14 (05:51 UTC, November 15), an internal routing component lost its connection and did not reconnect automatically. As a result, a substantial portion of TTS requests in our Asia Pacific (Mumbai) region failed or responded slowly. We identified the issue and restored service at 22:03 PT (06:03 UTC), and error rates and latency have been normal since. We’re improving how this component recovers more gracefully from connection failures in the future.