you have to configure via API on /credential endpoint, then use the custom-voice configuration provided from the docs page