same voice providers, etc? Is there any background noise? Maybe review the logs and see if the transcriber has issues or what was returned from the model.