I haven't seen this yet, but I imagine it would involve running at least "voice2json record-command | voice2json transcribe-wav | jq .text". This will record a single command (until silence), and output the text transcription.