1.8.16 moves Local Whisper to the official Whisper.net CUDA Windows runtime package so CUDA is discoverable through the app dependency graph, keeps CPU fallback disabled, stages downloaded CUDA dependencies beside the CUDA runtime, and shows the app version in the header and About tab.
\n1.8.15 fixes the NVIDIA-only loader path by pointing Whisper directly at the prepared CUDA whisper.dll while still refusing CPU runtime fallback.
\n1.8.14 makes Local Whisper NVIDIA-only: the installed CUDA runtime pack is copied into the app native runtime folder at startup, Whisper is forced to request CUDA with no CPU fallback, and the runtime status now fails if CPU loads instead of falsely reporting CUDA availability.
\n1.8.13 is a local-only caption-engine reset: downloaded Large v3 is preferred as the primary live/history model, live captions remain provisional, history is emitted from utterance passes instead of stable-live shortcuts, draft transcript fallback waits longer for the accurate pass, and hardcoded mishear corrections were removed so recognition quality comes from Whisper instead of phrase swaps.
\n1.8.12 restores the proper hybrid Whisper pipeline: Base q5_1 stays on the instant live lane while the selected model, such as Large v3, runs a context-aware final polish pass with whitelist vocabulary prompting and faster draft replacement during continuous speech.
\n1.8.11 adds a focused live-caption correction pass for common homophone/context errors seen in the field, including days old/daze old and months old/moths sold phrasing, before display/history/transcript filtering.
\n1.8.10 fixes live-to-history handoff by promoting a stable processed live caption into history after a short delay instead of waiting for the next utterance, while clearing that text from the live area once it moves.
\n1.8.9 tunes live recognition by moving the realtime lane from Tiny to Base, using sane 0.9-second live windows, filtering Whisper non-speech tokens such as [blank audio], and giving the live decoder enough tokens to form short phrases.
\n1.8.8 makes Local Whisper live-first during active captioning: the app explicitly disables large-model polish in the live run, requires only the Tiny realtime model to start, lowers the audio chunk timing, and keeps live/history output on the fastest local lane.
\n1.8.7 keeps the large Whisper polish pass from competing with live captions: final polish is queued and only runs after the live lane has been quiet, with a small bounded queue so accuracy work cannot build up and steal GPU time during active speech.
\n1.8.6 restores the snappy live-caption pipeline by separating the realtime lane from the polish lane: Local Whisper now auto-prepares a Tiny q5_1 model for live captions/history drafts while the selected model remains available for final accuracy in the background.
\n1.8.5 decouples historical captions from the polished Whisper final pass by immediately promoting realtime captions into history, then replacing the draft with the polished final when it arrives; draft text is committed if the final pass never returns. It also shortens local Whisper finalization timing so local captions react faster during normal speech pauses.
\n1.8.4 lets the Room Agent reclaim a stale Event Live Captions process that is still holding the configured control port before it binds, preventing yesterday's abandoned listener from blocking Control Center pairing/connectivity.
\n1.8.3 hardens local and remote stop/start handling so caption shutdown is idempotent, time-boxed, and locally logged instead of crashing; preloads and keeps the selected Local Whisper model warm; gives realtime captions a dedicated latest-audio-wins quick lane; holds final captions in the live area longer; and adds a Ten Fifty Entertainment branding preset.
\n1.8.2 fixes Control Center room pairing against the Room Agent by supporting chunked HTTP request bodies, trimming PIN input before pairing, and sending pairing JSON with a stable content-length body from the embedded room-control client.
\n1.8.1 fixes Room Agent PIN pairing from the standalone Control Center by accepting the pairing request JSON reliably and returning clearer pairing errors when the request is malformed.
\n1.8.0 adds Room Agent network discovery and pairing: rooms advertise their programmed room name, venue/area, hostname, IP address, and version on the LAN; controllers can add rooms through discovery, manual IP/hostname entry, or temporary PIN pairing; paired rooms can be refreshed and controlled remotely from the Files settings panel.
\n1.7.0 adds the first Room Agent foundation for local network control center work: room identity settings, a token-protected local HTTP agent, status/snapshot endpoints, preview PNG output, and remote commands for start, stop, clear, save, output, and device refresh.
\n1.6.0 adds NVIDIA GPU detection on startup, a first-run CUDA runtime install prompt, optional automatic CUDA runtime download for NVIDIA systems, and clearer GPU/runtime status in the operator UI while keeping the base installer small.
\n1.5.0 moves the CUDA 13 acceleration payload out of the base installer and into a verified runtime pack downloaded from the ELC update feed. The app now reports CUDA runtime status in Controls, can install the external runtime pack on demand, and still falls back to the bundled CPU Whisper runtime when CUDA is not installed.
\n1.4.0 polishes license activation with clearer key-vs-ID guidance, clipboard paste extraction for ELC license keys, local validation before server calls, and persistent license clearing.
\n1.3.0 adds desktop license activation/validation against the ELC licensing service, per-machine ID binding with offline grace, the packaged ELC SVG/PNG logo in the app header, transparent collapsed menu icons, and portable branding presets with relative logo assets.
\n1.2.0 adds custom ELC collapsed menu icons, a lock for collapsed side menus, smoother non-blocking local Whisper interim captions, live caption handoff cleanup, and the first ELC licensing server/admin page foundation for eventliveproductions.com deployment.
\n1.1.0 keeps the 0.1.18-style operator layout while adding fixed settings tab positions, clearer Start/Stop/Clear action colors, portable branding files, and full portable ELC settings files that can launch the app with saved configuration.
\n1.0.0 starts the clean Event Live Captions application line with ELC package IDs, ELC folders, EventLiveCaptions executable naming, Toro Ventures Group-inspired gunmetal and gold branding defaults, branding settings, local captioning, CUDA-first Whisper support, output layout controls, audio source controls, and update-ready Velopack packaging.
"}]}