For the exact same event.transcript, console.log prints correct Hebrew, while logger.info outputs mojibake/corrupted characters. The UserInputTranscribed event exposes the transcript directly, so this seems isolated to logging rather than STT.
Thanks for the detailed report, the repro with both loggers side by side made this quick to pin down.
Good news: your transcript isn’t being corrupted. The event.transcript string is correct all the way through; what differs is how each call gets bytes onto the Windows console.
This seems to just be an issue in the dev console on Windows. You are not seeing it on a prod env right?