개인정보

Who is responsible

MP3 to Text is a Chrome extension published by Individual Entrepreneur Aleksandr Gerasimov (Tbilisi, Georgia), part of Bear Tools. For anything in this policy — including a request to see, correct or delete your data — write to apps@beartools.dev.

This policy covers the MP3 to Text extension and the pages on this website that serve it. The wider Bear Tools privacy notice covers the rest.

The short version

The extension transcribes with one of two engines, and which one you pick decides where your audio goes. The panel names the selected engine before every run, and the choice is made per file. Out of the box the cloud engine is selected.

On the on-device engine, no audio, no transcript and no file name leaves your computer. On the cloud engine, the recording is uploaded so a transcription provider can process it. This page describes both, because a privacy policy that only describes the flattering half is not one.

No account is needed to use the extension. Buying Pro takes an email address, which Paddle collects at checkout, and that address is how you sign in to Pro in another browser — with a code sent to it, never a password. Personal data is never sold, and there is no advertising and no cross-site tracking anywhere in the product.

The on-device engine

Local transcription decodes and transcribes audio inside your browser. The audio file, the transcript and everything derived from them stay in your browser profile on your machine.

One network request is involved, and only the first time: the speech model and the speech-detection model are downloaded from Hugging Face (huggingface.co and its content delivery hosts) and then cached. Like any download, that request tells Hugging Face your IP address and browser — it carries nothing about your file, because at that point no file has been opened. After the download the engine works with no network at all.

Its free minutes are counted by the extension itself, in your browser's storage. Our service keeps no usage record for on-device runs and never learns how much you have transcribed this way — the only trace that leaves your computer is the bucketed usage analytics described below, and only while they are switched on.

The cloud engine (Pro Transcript)

The cloud engine is faster and more accurate, and it works by uploading the recording. It runs only for a file you started on that engine.

When you start a Pro run, the extension sends the recording itself, its file name, its size, its duration and its format, together with your installation identifier (see below). The audio is uploaded straight from your browser to private object storage using a single-purpose signed URL; it is then fetched by the transcription provider, transcribed, and the text is returned to the panel.

Switch the engine to the on-device model and none of this happens: nothing is uploaded and no request is made at all.

• The recording is deleted from storage as soon as the provider returns a result, and also when a run fails, is cancelled, or is abandoned.
• The job record — which holds the file name, the installation identifier and the finished transcript — is kept for up to 24 hours so the panel can collect a result after a browser restart, and is then deleted automatically.
• A usage record is kept on our service for every Pro Transcript run — and only for those — so that a free or Pro allowance cannot be reset by clearing the browser. It holds the installation identifier — or, on Pro, the subscription it was counted against — a random job identifier, a number of seconds and timestamps. It holds no audio, no transcript and no file name, and it is kept for as long as the service runs.

The transcription provider

Speech recognition for the cloud engine is performed by Deepgram, Inc. (United States) on our behalf, under their terms and their own privacy commitments. The recording is transmitted over an encrypted connection and is sent for one purpose only: producing your transcript.

As is normal for machine-learning services, audio and transcripts handled by the provider may also be used by them to operate, secure and improve their speech recognition, in accordance with the agreement in place between us and the terms they publish. If that matters for your material — recordings covered by professional confidentiality, for example — use the on-device engine, which never sends anything anywhere.

What is stored on your own computer

Most of what the extension holds never leaves your machine, and you can clear it yourself at any time from Settings → Stored data.

• Staged audio, transcripts, your edits and job state, in the browser profile. Abandoned jobs are deleted automatically after seven days.
• Your saved transcripts (the library), until you clear them.
• Your settings, including the engine you chose and the analytics switch.
• How many on-device minutes you have used, including any minutes granted on top of the free allowance.
• The downloaded speech models, which can be cleared separately with "Free up cached engine data".
• Removing the extension from Chrome deletes all of it, including the installation identifier.

The installation identifier

The extension generates one random identifier when it is installed (a UUID from the browser's own random source). It is not derived from you, your device, your browser profile or any hardware property, it is not a fingerprint, and it says nothing about who you are.

It has three jobs: it is what our service counts the free Pro Transcript minutes against, it is the identifier attached to analytics events, and, if you buy Pro, it is what the subscription is linked to, so the extension knows Pro is on. It stays the same until you remove the extension or reinstall it; a reinstall gets a new one, and Pro reaches it when you sign in there. Note that "Delete local data" deliberately does not reset it — that button clears your recordings and transcripts, not your remaining minutes.

Buying Pro

Pro is sold through Paddle.com, which acts as Merchant of Record: Paddle runs the checkout, takes the payment, issues the receipt and invoice, and handles tax. At checkout Paddle collects your email address, your country — and a postcode where tax depends on it — and your payment details, under its own privacy policy. Card details never reach us.

• Showing prices: when the panel shows the Pro plans, our service asks Paddle for the price and tax in your country, passing on your IP address so Paddle can work the country out. The answer is held in memory for up to an hour and nothing from it is written to our database. If Paddle does not sell to your country, we keep a note that this installation asked, with the address cut to its network (/24 for IPv4, /48 for IPv6), so we can judge whether another payment provider would be worth adding.
• What we keep after a purchase: your email address, Paddle's customer and subscription identifiers, whether you agreed at checkout to hear from us, the plan, its status and billing dates, and the notifications Paddle sends about it — payments, renewals, cancellations and refunds. It is linked to the installation identifier the purchase was started from, and to any installation you sign in from.
• Signing in: when you ask for a code, we keep the address, a keyed hash of the code and when it was sent, for up to two hours — that is what limits how many codes one address can be sent. The code itself is only in the email. An address with no subscription is sent a note saying so instead, and the extension shows the same thing either way, so nobody can use it to find out who is a customer.
• A signed-in browser holds a session token in Chrome's sync storage, so it follows your Chrome profile. We keep a hash of it and when it was last used; signing out in Settings revokes it.
• We use your email address to run the subscription, to send sign-in codes you ask for, and to answer you. We do not send marketing email unless you agreed to it at checkout.
• Purchase records are kept for as long as the subscription exists and afterwards for as long as accounting and tax law require.

Usage analytics

Usage analytics are on by default and can be switched off at any time in Settings → Privacy. Switching them off stops events being created and deletes anything still queued. They are processed by PostHog on its European infrastructure.

They are a small request wrapper rather than a full analytics SDK: no autocapture, no session replay, no page views, no person profiles, no advertising identifiers.

• An event carries the installation identifier, a random per-job identifier, the extension version, and bucketed values — a duration band rather than a duration, an error code rather than an error message, "pro" or "transcript" rather than a model name.
• Content is structurally excluded: a deny-list drops any property named like a file name, path, transcript, email, user agent or device, and every remaining string is truncated to 64 characters, so a coding mistake cannot leak text.
• Like any web request, sending an event reveals your IP address to PostHog. It is not used to build a profile and is not stored as part of the event.
• On installation the extension records your browser's language, without the regional part — "de", never "de-AT" — to decide which languages are worth translating next.
• Subscription events — started, renewed, payment failed, cancelled, refunded — are sent to PostHog by our service, from what Paddle tells it, under the same installation identifier. They carry the plan and nothing about you. They follow the same switch: the extension tells our service whether analytics are on, and nothing is sent for an installation that has them off.

When you write to us

Three places in the product ask for something you type, and all three are optional.

• The Pro waitlist, shown where Pro cannot be bought yet: your email address, the offer you were shown, and a one-way hash of the installation identifier. It is stored so we can tell you when paid minutes open, and a copy is emailed to us so we can answer you.
• The feedback and uninstall forms, and the rating prompt in the panel: your score, your message, your email address if you leave one, your interface language, the extension version, your browser's user-agent string, and a salted daily hash of your IP address used only to rate-limit the form. A bare score with no message is stored but not emailed.
• Support email: kept for as long as we may need it to answer follow-ups, and no longer.

Who processes data for us

These are the only third parties involved, and each one receives only what its job needs:

• Deepgram, Inc. (United States) — speech recognition for the cloud engine.
• Railway (object storage, the transcription API and its database) — holds the upload and the job record for the windows described above, and the usage and purchase records.
• PostHog (European Union) — usage analytics, when they are switched on.
• Hugging Face — hosts the on-device speech models that the extension downloads.
• Supabase — stores waitlist and feedback rows.
• Resend — delivers sign-in codes and the notification emails those forms produce.
• Vercel — hosts this website and the feedback endpoint.
• Paddle.com (United Kingdom) — Merchant of Record for Pro: checkout, payment, invoicing, tax, and the customer portal where you manage a subscription. It also prices the plans for your country.

Permissions, and why each one exists

The extension asks for the narrowest set that makes it work, and for no host permissions at all — it cannot read the pages you visit, and cannot contact any server other than the ones named above.

• Side panel — the only interface the extension has.
• Offscreen — hosts the transcription worker so a job survives the panel being closed.
• Storage — job state and your settings.
• Notifications — tells you when a background job finished or failed. They never carry a file name or a word of transcript.
• Alarms — recovers an interrupted job and runs the seven-day cleanup.

How the data is used, and how it is not

Your use of MP3 to Text is subject to the Chrome Web Store User Data Policy, including its Limited Use requirements. Concretely, and in plain terms:

• Data is used only to provide and improve the features described on this page.
• Personal data is never sold, and never transferred for advertising, credit-scoring or lending purposes.
• Data is transferred only to the processors listed above, for those purposes, or where the law requires it.
• No human reads your transcripts. Nobody at Bear Tools has an interface that shows them.
• There is no advertising in the product and no cross-site tracking of any kind.

International transfers

We operate from Georgia, and the processors above operate in the European Union and the United States. Where data from the European Economic Area or the United Kingdom is transferred outside it, that transfer rests on the standard contractual clauses or an equivalent safeguard in the relevant processor's terms.

Your rights

You can ask for a copy of what is held about you, for it to be corrected, for it to be deleted, or for it to be exported. Write to apps@beartools.dev and we will answer within 30 days.

Most of it you can also do yourself and immediately: Settings → Stored data clears your recordings and transcripts, the analytics switch stops and deletes the event queue, and removing the extension deletes everything it kept locally, identifier included.

If you are in the European Economic Area or the United Kingdom you may also complain to your national data protection authority. If you are in California you have equivalent rights under the CCPA, including the right not to be discriminated against for using them — and we do not sell personal data.

Children

The extension is not directed at children and is not intended for anyone under 16. We do not knowingly collect data from them.

Changes to this policy

This policy was last updated on 1 October 2026. If the data the product handles changes, this page changes with it before the release that changes it ships, and the date above moves.