Voice recordings are turned to text via in-browser Speech-to-Text (STT the opposite of TTS) and then piped / sent to Grok as text. These are privately sent via my API key to Grok. I don't hold on to them at all except for sending them over the wire to Grok. So storage duration on our side is as minimal as possible, typically just a few seconds while it is sent to Grok and a reply is formulated. However, one should be aware that the whole conversation is sent to Grok from within our lesson pages, so the whole chat persists for a session until you close out the tab or move to another lesson page, at which point it is reset.