Microphones, browsers and privacy

Which browsers work

Every voice feature in Rhetoric, the teleprompter, freestyle and debate, relies on speech recognition to turn your voice into text. How well that works depends on your browser. Desktop Chrome or Edge on a laptop is the recommended setup.

• Chrome and Edge (desktop): full support for the enhanced tracker, the browser engine, and live voice debate. This is the recommended setup.
• Safari (Mac): the browser engine works. The enhanced tracker connects when a secure context, AudioWorklet, and a microphone are available.
• Firefox: Firefox does not have a built-in speech recognition service, so the browser engine and the public demo are not available there. The enhanced tracker works on Firefox for paid plans, because Firefox supports AudioWorklet and secure contexts. Without a paid plan, timed scroll is the fallback.
• Phones and tablets: you can browse, subscribe, and read your library. For voice practice, a laptop with Chrome or Edge gives more reliable tracking.

Sessions are labeled before they start, so you always know which engine is listening.

Three ways to listen

Rhetoric uses three ways to listen, depending on your plan and your browser. The product names them on the prep screen before a session begins.

• Enhanced tracker (paid plans): a managed speech engine with smoother word timing, confidence and recovery. Your browser sends microphone audio directly to the speech provider. Rhetoric does not receive the audio.
• Browser engine (the public demo, and the fallback): your browser's own speech recognition service. Audio is processed by that service, for example Chrome's or Safari's, and may pass through their vendor's servers. Rhetoric does not receive it.
• Timed scroll (any browser, any plan): no microphone at all. The script scrolls at a steady pace you set, between 90 and 180 words per minute. It is not voice tracking and it does not pretend to be, but it keeps you rehearsing anywhere.

If the enhanced engine cannot start, the session falls back to the browser engine and says so. The same honest label applies. If neither engine is available, timed scroll is always there.

Granting microphone access

The first time you start a voice session, your browser asks whether Rhetoric may use the microphone. Allow it once and the browser remembers. The prep screen before every session shows which engine will listen: the enhanced tracker, your browser's engine, or timed scroll when voice is not available.

1. 1Chrome or Edge: click the icon to the left of the address bar (a lock or sliders), find Microphone, switch it to Allow, and reload the page.
2. 2Safari: open Safari menu, then Settings, then Websites, then Microphone. Set the site to Allow and reload.
3. 3Firefox: click the permissions icon in the address bar and remove the microphone block. Reload and choose Allow when the browser asks again.

If the browser allows the mic but nothing happens, check your operating system. macOS System Settings, then Privacy and Security, then Microphone must permit the browser. On Windows, Settings, then Privacy, then Microphone. External mics: check which input device your OS has selected.

The public demo and the paid tracker

The public demo on the marketing site is a lightweight sample of voice tracking. It uses your browser's own speech recognition, runs without an account, and does not save sessions. It is a quick way to test your setup.

Paid rehearsal is a different product. Where a managed speech engine is used for smoother word timing, confidence or recovery, the product labels that clearly before the session starts. Sessions are saved to your history, and your workspace holds your library, scores and progress. The demo is a taste of one feature. Paid is the whole practice room.

Live voice debate

Premium adds live spoken debate: you speak, the opponent answers in a voice, and turns are taken by ear. The audio path uses a voice agent connection for the opponent's speech and a caption stream so your transcript appears while you talk. Audio is processed live to produce the transcript and is not stored by Rhetoric.

What Rhetoric keeps

Rhetoric never stores your raw audio. Voice is transcribed live, and the audio is then gone. In every case, whether the enhanced engine, the browser engine, or the trial relay, Rhetoric does not record or retain the audio itself.

What persists is the product you can see: your speeches, transcripts, practice metrics (pace, coverage, recoveries, scores), debate records, and account and subscription state. Support requests you send are kept too, with their retention described in the Privacy Policy.

• AI features (analysis, coaching, the debate opponent and judge) send the relevant text to the AI provider. Per its commercial terms, API data is not used to train models.
• Billing is handled by Stripe. Your card number never touches Rhetoric's servers.
• Usage analytics are tied to your account id, never your speech text. You can turn analytics off on the cookie preferences panel.
• If your subscription lapses, your library stays readable and exportable. Rhetoric never holds your writing hostage.

The free trial audio path

On the free trial, voice sessions travel through a relay that Rhetoric operates. The relay passes the audio to the speech provider so the session can be ended at its time limit. It does not record or keep the audio. Trial debates are limited to 60 seconds each and trial freestyle to one session.

Export and deletion

From Settings you can export everything you have made as a single JSON file: profile, subscription state, speeches, practice sessions, and support requests. If any part of the export cannot be produced, it fails loudly rather than handing you an incomplete file.

Individual sessions, including transcripts, can be deleted from their pages. Account deletion in Settings is self-serve and immediate. Your content, sessions, and support requests leave the live database. Billing records stay only as long as tax law requires. If you have an active subscription, cancel it in the billing portal first. The screen reminds you.

When something goes wrong

Most microphone and tracking problems have a short fix. These Help articles cover the common cases and walk through each one step by step.

• Allow or restore microphone access: browser-by-browser steps for a blocked or denied microphone, including the operating system check.
• Supported browsers, engines, and fallbacks: which engines work where, what the enhanced tracker is, and what timed scroll offers when recognition is not available.
• Troubleshoot teleprompter tracking and pace: what to try when the stage lags, stalls, or will not advance, and why names and invented words are the hardest for any recognizer.
• Your voice, your data: export and deletion: what exists in your account, how to export it, and how to delete individual items or everything.