I am using its voice model for mandarin with Azure transcriber, not bad