extempore.co.uk

Replace the browser Web Speech read‑aloud with an on‑device Kokoro‑82M TTS that generates audio locally (bundled into the Start download). The female voice maps to a British voice and the male voice to an American voice; audio is generated on the device so no cloud speech service is used.

Add 56 new prompts aimed at enterprise audiences across all categories (prompt pool increased from 317 to 373) so you can practice board, procurement and large‑scale architecture scenarios.

Topics and all session screens now use an audience/customer-type (non-technical, SMB, mid-market, enterprise, internal IT) instead of the previous difficulty bands, and you can pick the audience when creating a session.

Setup now offers a TTS voice picker (Auto or explicit voice) and a Test button so you can preview the read‑aloud voice before a session, and the chosen voice is persisted.

You can now tap a history entry to open its full scorecard and restore the original drill so Retry/Next continue from the same topic.

Added a new Service Operations / Support & Incident Management category and many new prompts and resources so the topic database now contains 317 curated prompts across more IT areas.

Prep now shows a 'Preparing' spinner while the voice is being generated, reveals the question when playback actually starts, and then runs the prep countdown; replay and the start button are disabled while generation is in progress to avoid stuck or silent states.

The TTS model is permission‑gated and begins downloading only when you press Start; the model is warmed during that download so the first read‑aloud plays without the previous cold‑start delay.

Resume the browser AudioContext from the Start click (a user gesture) so on‑device read‑aloud is not blocked by autoplay restrictions and is audible on first play.

Update the public site to a lighter, professional 'PatchPen' style theme (ink + accent colours, no gradients) for a cleaner, more professional look.

Expand the audience selector with customer size‑band options (for example small 0–50 and large enterprise 5k+) so prompts better match the intended audience.

Updated the product name, homepage copy, UI labels and docs to 'Extempore' and refreshed the homepage messaging and feature highlights.

The homepage now includes meta description, keywords, canonical URL, Open Graph and Twitter tags, an og image and JSON‑LD so link previews and search listing information are improved.

The intro/home page now surfaces a short 'Roadmap coming soon' note in the features area to signal upcoming scenario packs and progress tracking.

The app now ranks and auto‑picks the best English neural voices more reliably, warms the chosen voice on startup so Speaking mode is snappier, and reads only the question (not the scenario) when speaking.

Your preferred TTS voice and the Speaking/Text mode toggle are stored in settings so these preferences persist between visits.

If the browser's live-caption service errors or is unsupported, the captions panel now shows a helpful notice so you know transcription will still run locally when you stop.

The app now cancels any in‑progress TTS before recording and when leaving prep so the spoken prompt won't be included in your recorded answer.

Live captions now attempt automatic restarts where possible and surface fatal speech-recognition errors so captions don't silently stop working mid-session.

History and settings storage keys were renamed from 'soundcheck:history' / 'soundcheck:settings' to 'extempore:history' / 'extempore:settings', so any sessions or settings saved under the old keys will not be migrated and will not appear after the update.

The time selection was removed and new sessions always use a 60 second limit (Start now shows '60s'); users who relied on choosing different talk times will no longer be able to change the session length.