v0.99.54

Google's Gemini voices on calls, a never-play check on every list the DJ offers from, and the rest of the approved station reads.

• Google (Gemini) voices. Pick the google-gemini-cloud.json adapter under Voice and calls speak with the same thirty Gemini voices SUB/WAVE's own engine uses, on the Google key your Gemini brain already has. A DJ the station voices through Gemini keeps that exact voice on calls. First audio arrives in about a second (measured on a live key).
• The DJ stops offering never-play tracks. Similar tracks, your playlists and the play history now carry the station's never-play mark, so the DJ leaves those tracks out instead of promising one and being refused. A playlist queues without them and says how many it skipped.
• "Was that my request?" names who asked. The play log's requester never reached the DJ because of a field-name mismatch; it does now.
• Where listeners tune in from: a new permission, off by default. It adds one line to the DJ's briefing naming the countries in the station's last-day rollup with at least three listeners each.
• A restarting station says so. The settings page's station checks now say "the station is starting up" instead of printing a raw 503.
• Who's listening now and Untagged songs on the Diagnostics page: the station's own admin reads, loaded when you press Load.
• Fixes: the after-call transcript button no longer gets cut off on a narrow phone, because the rating moves to the next line. The installed app in a laptop-sized window shows the queue instead of clipping it. A record with no art no longer spills its initials outside the cover on the desktop card.
• Under the hood: the voice module was split along its planned seam, and the upstream drift check now reports new SUB/WAVE voice engines.