Support & FAQ

Google Chrome (desktop). TrainScription is a Manifest V3 Chrome Extension and requires a Chromium-based browser that supports the Chrome Extension APIs. It does not currently support Firefox, Safari, or mobile browsers.

Browser Tab mode works with any browser-based meeting platform: Google Meet, Microsoft Teams (web), Zoom (web browser), Webex, Discord, and any other platform that runs in a Chrome tab and uses browser audio. If the meeting is in a tab, it works.

Full Desktop mode captures all system audio from your machine — so it works with the Teams desktop app, Zoom desktop client, Discord app, Slack, and anything else making sound on your machine. Not just browser tabs. Select your screen from the OS picker and enable "Also share system audio" when prompted. Free tier sessions run with the standard 15-minute cap and other free-tier limits — Pro removes those.

Only for the initial AI model download on first use. After the model is cached on your device, transcription works fully offline. The extension has no cloud dependency for transcription.

Not by default — Chrome extensions are blocked from incognito unless you explicitly allow them. To enable TrainScription in incognito: open chrome://extensions, find TrainScription, and turn on "Allow in Incognito."

Full Desktop mode captures all system audio — your microphone plus everything your machine is playing — so it works with any desktop app, not just browser tabs. It's available on Free and Pro — Free sessions are subject to the 15-minute cap.

To use it: select Full Desktop in the capture mode toggle in the popup, then click Begin Capture. Chrome will show an OS screen picker. Select Entire Screen (not a specific window) and make sure "Also share system audio" is toggled on — this is the critical step. Confirm. A small capture window will open briefly and then minimize to your taskbar. Transcription begins immediately.

If you don't see the capture window in your taskbar, the session didn't start — check that you enabled system audio sharing and try again.

Almost always this means "Also share system audio" was not toggled on in the OS picker. Chrome requires this to be explicitly enabled — it's off by default. Cancel the current session, click Begin Capture again, select Entire Screen, and make sure the system audio toggle is switched on before confirming. On Windows the toggle is labeled "Share system audio." On Chrome OS it's labeled "Share audio."

If you only see a window picker (not a screen picker), cancel and try again — Full Desktop mode requires Entire Screen to be selected, not an individual app window, for system audio to be available.

Microphone access is required to capture your local audio channel so both sides of a conversation can be transcribed separately. The permission is only active during a capture session you explicitly start. It is not used for any other purpose — no background listening, no passive monitoring.

Audio is processed locally in 5-second volatile chunks in browser memory and discarded immediately after transcription. Never written to disk. Never transmitted anywhere. There are no TrainScription servers — there is no TrainScription infrastructure of any kind.

TrainScription runs a real AI model on your machine, so you will see CPU activity during sessions — that's expected and normal. On modern hardware it's comfortable. On older or heavily loaded machines you may feel it more. The extension gives you three processing tiers in the popup: Lightweight uses the fewest resources with a modest accuracy tradeoff, Balanced is the default and works well for most hardware, and High Quality produces the sharpest output at the highest resource cost. If you're on older hardware or running a long session, Lightweight is the right choice.

Yes — tab audio capture picks up any browser tab audio, so it works on YouTube, podcasts, recorded video, or any other audio playing in Chrome. TrainScription was designed for live conversations, but the capability extends to any browser-based audio you want a text record of. For longer media sessions, consider switching to Lightweight mode in the popup to reduce sustained CPU load.

Yes — and it's a surprisingly practical use case. Open the Studio and start a microphone capture session. TrainScription will transcribe everything picked up by your mic, including other voices in the room. Since there's no diarization, all speech is attributed to ME, but the transcript is still a complete, timestamped record of the conversation. From there, take the exported .txt to any AI — ChatGPT, Claude, Gemini — give it the context that it was a two-person meeting, and you get summaries, action items, or key decisions in seconds. No cloud meeting platform required. No screen share. Just open the Studio, hit capture, and let it run.

The Whisper AI model is excellent at general language but consistently struggles with proper names, product names, jargon, and specialized terminology. This is expected behavior across all AI transcription tools. The Phonetic Brain is built specifically to fix this — highlight any wrong word in the transcript and assign the correct spelling. The Brain corrects that word on every future transcript automatically, before you ever see it.

Mint-highlighted words were corrected by the Phonetic Brain in real time. Whisper produced an incorrect transcription; the Brain caught it and substituted the correct spelling before it was written to the transcript. The highlight shows you the Brain is actively working — and shows exactly which words it has touched.

Yes. Transcription runs in the extension's background service worker. Closing or minimizing the popup does not stop it. Your session continues and all segments are saved to Recovery automatically. Reopen the popup or Studio view at any time to see the latest transcript.

Silence Auto-Stop automatically ends a transcription session after a configurable period of sustained silence — so you never have to worry about leaving a session running indefinitely when a meeting ends.

Configure it in Studio → Settings. Options: 5 min, 10 min, 15 min, or Indefinite. Free tier: 5 and 10 minutes only. Pro: all four options including Indefinite.

At 1 minute before the threshold, a gold warning banner appears in the Studio Live view with Keep Transcribing and Stop Now buttons. A system notification also fires. If audio resumes before the threshold, the timer resets automatically — Keep Transcribing is only needed if the silence is real but you want the session to continue anyway. The transcript is always saved to Recovery when auto-stop fires.

Transcript history is stored in your Chrome profile's extension storage — export any sessions you want to keep before wiping or switching machines. Your Phonetic Brain vocabulary can be exported as a .train.json backup and reimported after reinstall. Your Pro license is tied to your email and carries over automatically — just sign in on the new device. The easiest path for Pro users: Recovery → Backup Everything on the old machine, then Restore from Backup on the new one — brings back every session and your full Brain in one step. See the Bulk Export, Backup & Restore section below for detail.

Yes — every segment in Recovery has a .srt button alongside the standard .txt button. SRT export produces a standard subtitle file with timestamps relative to the start of the session. This is useful both for meeting transcripts and for any browser-based video or media you've captured. Free and Pro.

Yes. Click the edit icon next to any session in Recovery to give it a custom name — "Weekly sync," "Client call," whatever makes sense to you. The name is saved immediately and replaces the default timestamp label. Leave it blank to revert to the default. Free and Pro.

Yes — this is separate from deleting a session yourself, which has always been available with a single click. Auto-Purge by Age, in Studio → Settings, lets you set Recovery to automatically clear out sessions once they pass a certain age: 30 days, 60 days, 90 days, or Never — Never is the default, so nothing is removed unless you turn this on. A daily background check compares each session's age against your chosen threshold and removes anything older, no manual cleanup required.

If a session matters and you never want it swept — regardless of how old it gets — click the lock icon next to that session in Recovery. A locked session is permanently exempt from Auto-Purge; age stops applying to it entirely. You can lock and unlock sessions at any time. This is the only thing that exempts a session from the sweep — there's no separate "keep the newest N" setting, and locking has no effect on the Free tier's 3-session limit, which is a different, count-based rule. Auto-Purge is ungated — the same options are available on Free and Pro.

Session Tags are the answer. Use the tag control on any session card in Recovery to assign your own labels — "Client Calls," "Weekly Sync," "RPG," whatever makes sense for your workflow. A row of tag chips appears above the session list; clicking one filters the list to every session carrying that tag. Clicking it again clears the filter.

Tags use autocomplete drawn from every tag you've already created, so if you tag a session "Client Calls" today, tomorrow's session surfaces the same spelling as a suggestion — no near-duplicates accumulating over time.

Bulk export is tag-aware: when a tag filter is active, the export button scopes to just the filtered sessions and labels the .zip accordingly, so you can package one category of sessions without downloading everything.

Session Tags are fully ungated — the same on Free and Pro. Tags are included in Backup & Restore, removed with the session when Auto-Purge sweeps it, and wiped by Privacy Nuke along with all other local data.

When you pause a channel — mic, tab, or both — TrainScription logs a timestamped marker line in the transcript: for example, [ME PAUSED] or [ALL RESUMED], using your configured channel labels. These markers document exactly when channels were deliberately silenced, so the record distinguishes an intentional pause from a natural silence gap. They appear in exports by default. If you want a clean export without them, toggle "Include pause markers in export" off in the Recovery section before downloading.

The fastest way: highlight the incorrect word in the live transcript or in any Recovery segment. A popover appears — type the correct spelling and confirm. That mapping is saved immediately and applies to every future transcript. You can also add terms manually in the Brain panel if you already know how Whisper tends to mangle a specific word.

Open the Brain panel and use the export function to download your vocabulary as a .train.json file. It's plain JSON — human-readable and editable outside the extension if needed. Import it on any other device or installation using the Brain import function (Pro feature). Keep it wherever you keep important files.

If you catch it immediately — a one-shot Undo appears in the correction toast right after you confirm a Brain fix. Tapping it reverses both the text change and the variant that was added to your Brain, as if the correction never happened. This is the fastest path for an accidental fix made moments ago.

If you notice it later — open the Brain panel, find the term, and expand it. You can remove individual phonetic variants or delete the whole term with a single click. No JSON editing required.

Yes — this is a Pro feature. Open Studio → Import, choose your file, and TrainScription runs it through the exact same on-device Whisper pipeline used for live sessions. Your file is read locally in your browser and processed entirely on your device — never uploaded anywhere, at any point.

Any audio or video format your browser can already decode — common audio formats, plus MP4, WebM, QuickTime, and MKV containers. If your browser can play it, Import can very likely process it. A small number of unusual codecs may not decode — if a file fails, that's the most likely reason.

Yes, automatically — exactly the same correction pipeline live sessions use. Your trained vocabulary fixes misfires in the imported transcript before you ever see it.

Import runs entirely in that browser tab, not in the background like a live session does. Closing or navigating away from the tab mid-import will stop it before it finishes — Chrome will warn you if you try. Once the import completes and saves, you're free to close the tab as normal.

The Import section is visible on Free so you know it exists, but running an import requires Pro.

In Recovery, click Export All Sessions. You'll get a single .zip containing every stored session — available on both Free and Pro. Free produces one .txt per segment; Pro produces one assembled file per session.

Export All Sessions is for reading or sharing — human-readable files, no extension needed to open them. Backup Everything (Pro) is for rebuilding — a machine-readable file meant to be fed back in through Restore from Backup, not read directly. Use Export to share or browse; use Backup if you're about to reinstall, switch machines, or just want real insurance.

On your old machine: Recovery → Backup Everything, save the resulting .zip somewhere you'll have access to later (cloud drive, USB, email to yourself). On the new machine: install TrainScription, sign back into Pro, then Recovery → Restore from Backup and select that file. Every session and your full Brain vocabulary come back. This is a Pro feature.

No — Restore is additive. Any session or Brain term that already exists is skipped, not overwritten or duplicated. It's safe to restore the same backup more than once.

No. All three are entirely local browser operations — reading what's already on your device, packaging it into a file, or reading a file back in. No network connection is involved at any point, in either direction.

Yes — the free BrainTrainer companion at trainscription.com/brain-train.html is built exactly for this. Type a word, hold the button, speak it a few times, and the app collects the mishears your device's speech recognition produces. Select the ones that look like genuine wrong guesses, export a .train.json file, and import it into TrainScription Pro. Your Brain knows the word before the first transcript line appears. It runs on Android and iOS, installs to your home screen, and works offline after the first visit — no extension required.

No. BrainTrainer uses your phone's built-in speech recognition — the browser's Web Speech API — not Whisper. The two engines have different failure signatures, so variants captured on the phone won't be an exact match to what Whisper produces. In practice the overlap is real and useful: testing showed both engines independently producing the same primary variants for brand compound words. Some Whisper-specific variants will still surface during live sessions and can be added then via highlight-to-fix. Think of BrainTrainer as the head start — live sessions do the refinement.

No — and this applies to both BrainTrainer and in-session training. Before adding a variant, ask: could someone actually say this phrase in the same conversation? If yes, skip it — adding it would cause the Brain to silently replace that phrase every time it appears legitimately. A concrete example: both Whisper and the Web Speech API produce "transcription" when trained on "TrainScription." That variant should not be added — users of this product will naturally say the word "transcription" in context, and replacing it would corrupt those lines. "Train description," on the other hand, is a nonsense phrase in any realistic conversation about this product — zero false-positive risk, safe to add immediately. When in doubt: if the variant could stand alone as a real sentence fragment that someone might say, leave it out.

In rough order of how completely a single training session covers them: real-but-unusual words — place names, client surnames, medication names — tend to have limited variant counts and can often be fully covered in one session; brand names made of real words — any SaaS product name, internal project name, or compound proper noun — both engines split and merge them predictably, usually producing 2–5 usable variants total; invented and fantasy words — character names, made-up terms, deep niche jargon — have high variant ceilings that may never be fully mapped, but the BrainTrainer plants the term early so live sessions build on something rather than starting from scratch. A good first word to train on either platform: TrainScription itself — it's a real compound word both engines will consistently get wrong, and adding it shows you immediately how the system works.

Copy Brain writes your trained vocabulary to your phone's clipboard. Paste Brain reads your laptop's clipboard. For this to work seamlessly, your phone's clipboard has to automatically appear on your laptop — which only happens on specific device combinations. But there are three ways to get there, and one of them works for everyone.

The note app bridge skips the file download entirely — you're just moving text through a synced app you already use. The Download .train.json → Restore path is the permanent floor and always works if anything else doesn't.

Free includes 15-minute sessions in both Browser Tab and Full Desktop mode, 3 Recovery sessions stored, 3 Phonetic Brain slots, real-time dual-channel transcription, pause controls, session naming, Session Tags, optional Auto-Purge by age with session locking, Silence Auto-Stop (5 and 10 min options), AI Summary Handoff with preset and custom prompt options, and per-segment .txt and .srt export. Pro ($9.99 once) removes the 15-minute cap from both capture modes — unlimited session length in Browser Tab and Full Desktop alike — plus unlimited Recovery storage, unlimited Brain slots, one-click full-session assembled export, Brain vocabulary import, Silence Auto-Stop with all options including Indefinite, AI Summary Handoff for full multi-segment sessions, Audio & Video File Import, Backup Everything & Restore, and cross-device access via email login.

Yes. $9.99 one time. No recurring charges. No monthly billing. The Pro license is yours permanently, including all future updates to the extension.

Log in with the same email address on each device. Pro status is tied to your email and verified via ExtPay. Install TrainScription on any Chrome browser, click Unlock Pro, sign in with your email, and your license is active.

Per-person — one individual across multiple devices they personally use. It is not a shared or team license. Each person in a team who wants Pro needs their own license.

Not in the current version. TrainScription is focused on English transcription accuracy. The underlying AI model has some multilingual capability, but that feature is not enabled in this release. Language support is on the roadmap for a future update.

TrainScription separates audio into two clean channels by architecture — your microphone (ME) and the meeting audio (THEM). Identifying individual speakers within the THEM channel — knowing which of three people on a call is talking — requires voice fingerprinting or speaker diarization, which is a significantly heavier processing task. Running that accurately on-device would require much more CPU and memory. Doing it reliably would mean sending audio to a cloud service built for diarization. Either path defeats the point. The ME/THEM split handles the most important distinction clearly and keeps everything local and lightweight.

Yes — via AI Summary Handoff. TrainScription produces the transcript: an accurate, corrected text record of what was said. When your session ends, go to Recovery, click Summarize, and choose how you want it handled.

Three preset styles are built in: Executive Summary, Action Items, and Full Notes. Each preset comes with a Modify option — clicking it opens the preset's full prompt text in an editable field so you can tweak it for that send without changing the preset itself. If none of the presets fit, select Custom and write your own instruction from scratch (up to 400 characters — enough to be specific, short enough not to crowd the transcript).

Whatever style you choose, your transcript and prompt are copied to your clipboard while your chosen AI platform opens in a new tab ready to receive them. Paste with Ctrl+V (Cmd+V on Mac) and you have a summary in seconds.

You configure which AI platform you trust in Settings — ChatGPT, Claude, Gemini, or Copilot. All options are off by default. You must explicitly enable one and explicitly click Send each time. Nothing is transmitted automatically, and TrainScription never sees your transcript after it leaves your device — it goes directly from your browser to the platform you chose.

This approach means your summary is produced by the AI you already use and trust, without TrainScription picking one for you or adding cloud infrastructure on our end.

No — and this distinction matters. Audio never leaves your device — that's fundamental to how TrainScription works and not something the Summary Handoff changes. AI Summary Handoff sends your transcript text only, and only when you explicitly click Send. The transcript is copied to your clipboard and your chosen platform opens in a new tab. TrainScription is not in the path between your clipboard and that platform — the text goes directly from your browser to wherever you paste it.

The platforms listed (ChatGPT, Claude, Gemini, Copilot) each have their own privacy policies governing what they do with content you submit to them. TrainScription has no visibility into or control over those policies. If this matters for your workflow, check the relevant platform's data handling terms before enabling the feature.

By design. Many transcription services store your audio on their servers as a fallback for their own processing errors — if the transcript is wrong, they can re-process the audio. TrainScription doesn't need that because the Phonetic Brain corrects errors permanently going forward, and the AI model runs locally with no server to store to anyway. The transcript is the deliverable. Your audio doesn't need to survive anywhere — and keeping it that way is exactly what makes this tool private.

The local-only architecture means no audio is transmitted to third-party servers through the transcription process, which is architecturally meaningful for conversations involving protected health information. However, HIPAA compliance depends on your full workflow, your organization's policies, and your IT and legal requirements — we can't make a blanket compliance claim. Evaluate it within your specific organizational context and consult your compliance team.