The same problem. Or I get empty metadata, otherwise the same issue if use synchronous speech recognition.