Privacy Policy
Effective November 5, 2026
AudioRinse is operated by Wikivec Inc. (“we”). This policy explains what we collect when you use AudioRinse, and what we never collect.
On the website, your recordings stay on your device
AudioRinse cleans audio inside your browser, on your own device. The files you open are read and processed locally and are never uploaded to our servers or anyone else's. We can't see, store, or share your recordings because we never receive them. The only things your browser downloads from us to do this are the program and the cleaning model. The exceptions are cloud settings you choose: Cloud lyrics, Cloud separation, Cloud speech repair and, on a phone, Cloud sheet music, below.
If you upgrade to Pro, the cleaned result is kept temporarily in your own browser's storage (IndexedDB) so it's still there when you return from checkout. It never leaves your device and is deleted once restored, or after two hours.
The interview transcriber works the same way: speech recognition and speaker detection run in your browser, and the recording and its transcript are never uploaded. Its models are downloaded once from Hugging Face, which sees your IP address and which public model file you downloaded, never your audio. Transcripts are saved only in your browser's storage on that device, where you can delete them. The transcription data-flow statement lists every request it makes. The other transcription and subtitle pages, such as transcribe audio and add subtitles to a video, are the same transcriber; a video with subtitles burned in is made in your browser too. “Analyse in Claude”, under a finished transcript, copies it to your clipboard (or saves it as a file) and opens a new Claude chat; nothing is sent from our page. If you paste it there and send it, Anthropic receives it under your Claude account's terms.
The vocal remover and stem splitter work the same way: the separation model runs in your browser and the song is never uploaded. The model is downloaded once from Hugging Face, which sees your IP address and which public model file you downloaded, never your audio; the browser keeps it, with a note of how fast your graphics chip ran a short speed test, so the next song starts at once.
The sheet-music tool works the same way: the piano check, the transcription and the rhythm models run in your browser, and the recording and its score are never uploaded. The models are downloaded once from Hugging Face, which sees your IP address and which public model file you downloaded, never your audio; the browser keeps them, so the next recording starts at once. If you upgrade to Pro, the score (not the recording's sound) is kept in your browser's storage across checkout, as above.
The BPM and key finder, tap tempo and the Camelot wheel download nothing and send nothing: the tempo and key are worked out in your browser from the song on your device. The pitch changer, key changer, speed changer and 432 Hz converter work the same way: the song is changed in your browser and never uploaded.
The voice recorder uses your microphone only while you record, once your browser has asked you. The recording is made in your browser and never uploaded. As you record it is saved in your browser's storage on that device (IndexedDB), so a closed tab doesn't lose it; it stays there until you start a new recording or delete it on the page.
Recent files. When you download a result, the file you started from is kept in your browser's storage on that device (IndexedDB), so you can open it again from the tool or your account page: the last 10, up to 1 GB. It is never uploaded and isn't part of your account. Remove a file, clear them all, or turn this off on the account page.
Cloud lyrics
On the song to lyrics page, signed-in accounts have a Cloud setting, which is off unless you pick it. With Cloud, your browser makes a small mono copy of the song and sends it to our server, which passes it to Google's Gemini API to write the lyrics and sends them back to your browser. We don't store the song: it isn't written to disk or to any database, and our logs record only its length and size. Under Google's Gemini API terms for paid use, Google doesn't use it to improve its products, and keeps requests for a limited time to detect abuse. The lyrics are saved only in your browser, like the other transcripts. The page's Fast and Accurate settings never upload the song.
We count how many songs you send each month (a free account includes 3, Pro 100), and how many free songs are sent each day, all accounts together. We record the counts, not the song.
Cloud separation
On the pages that split a song in two (the vocal remover and vocal isolator, the music and voice removers for video, the karaoke maker, the acapella extractor and the instrumental maker), Pro has a Cloud setting, which is off unless you pick it. With Cloud, your browser makes an AAC copy of the song (only its sound, for a video) and uploads it to our private file storage (Vercel Blob). Replicate, the GPU host, reads it from there through a link that works for one hour and separates it with the same open-source model the Best setting runs in your browser. Your browser then downloads the vocals from Replicate, and the rest is worked out in your browser. We delete the upload as soon as the separation ends (or after a few hours if something fails), and Replicate deletes its copy of the song and of the results within an hour. Our logs record the song's length and size, not the song. The Standard and Best settings never upload the song.
We count how many songs you separate in the cloud each month (Pro includes 100). We record the count, not the song.
Cloud speech repair
The de-clipper works in your browser. For speech it can't repair, Pro has a cloud repair under the result, which runs only when you press its button, for that one file. Your browser then makes a mono copy of the recording's sound (not a video's picture) and uploads it to our private file storage (Vercel Blob). Replicate, the GPU host, reads it from there through a link that works for one hour and repairs the voice with NVIDIA's open RE-USE model. Your browser downloads the repaired voice from Replicate and puts it back into its own copy of the recording. We delete the upload as soon as the repair ends (or after a few hours if something fails), and Replicate deletes its copy and the result within an hour. Our logs record the recording's length and size, not the recording.
We count how many recordings you repair in the cloud each month (Pro includes 100). We record the count, not the recording.
Cloud sheet music
On the sheet-music pages (audio to sheet music and the others), a phone is offered the cloud, because a phone browser can't hold the models. It's off unless you pick it, and then remembered on that phone. With it, your browser makes a small mono MP3 copy of the recording and sends it to our server, which writes the score and sends it back. We don't store the recording: it isn't written to disk or to any database, and our logs record only its length, the number of notes and the time taken. The score (the notes, no sound) is kept on our server for an hour so your phone can collect it if it was locked or reloaded, then deleted. We count how many cloud scores are made each day, all together, and limit how many one connection can make an hour; we record the counts, not the recordings.
AudioRinse in Claude
The Claude connector (AudioRinse in Claude) works differently, because Claude can't run the cleaning model in your browser. When you give it a recording, through its private upload link or a link you share, the file is cleaned on our servers (hosted by Vercel) by the same open-source model the website uses. Subtitles are made the same way: the speech is transcribed on our servers by Whisper, the open-source model the website runs in your browser. It is not sent to any other AI service.
- An uploaded recording is deleted as soon as it has been cleaned, or after 24 hours if it never is. A recording you make subtitles for is kept for those 24 hours, so the captions can be put into the video afterwards.
- The cleaned file and the two previews on the result page, subtitle files and captioned videos are deleted after 24 hours.
- The record of a job (file name, chosen options and the short report, no audio) is deleted about two hours after its files. For subtitles it also holds the transcript, and for burned-in captions the subtitle text you sent, so they go at the same time. Until then your account page lists your files.
- We count how many minutes you clean or subtitle each month, like the download count below.
- We never listen to your recordings, use them to train models, or identify anyone from their voice.
The connector receives only the file link or upload id and the options Claude sends it. It can't read your conversations, memory or other files.
What we do collect
- Account details if you create an account: your email address and how you sign in.
- Plan and billing status: whether you're on Free or Pro, when your subscription renews, its currency, and the day passes you've bought (when, for how long, and what you paid).
- Export counts: how many files you've downloaded this month, so we can apply the free limit. We record the count, not the file.
- Ratings, if you rate a result: the tool, good or not, what went wrong if you write it, and how the tool ran (the model, whether it used the graphics chip, the recording's length range), with your account or device identifier. Never the file, its name or anything in it.
- Why you cancelled, if you answer the question on Stripe's cancellation page.
- Your tools and settings: the tools you star, the ones you used lately, what you said you use AudioRinse for, each tool's settings (the strength, the format, the transcription language) and the presets and chains you save. They're kept in your browser, and with an account also in your account so they follow you to your other devices. Tool names and settings only, never a file or its name.
- Votes and suggestions on the next tools page: the tools you vote for (with your account), and a tool you suggest (with your account or device identifier). We read suggestions; they're never shown on the site.
- Invites: your invite link's code, which accounts signed up through it, and the free months it earned, so we can give them. A friend's name or email is never shown to you.
- Email about new tools, only if you ask for it: when you did. Every one has a one-click unsubscribe, and your account page turns it off.
- Tool counts: when a tool finishes a job, or you star or unstar one, which tool and on which day, to show which tools are popular. Nothing about the file. So that each device counts once a day, we keep a one-way code made from your IP address and browser for two days; it can't be turned back into either.
- A device identifier in a cookie (
ar_did), so the free limit works before you sign in. - IP address, to stop abuse of the download limit and the rating form, kept by our rate limiter for at most 24 hours, and in our host's standard request logs.
We don't sell personal information, and we don't use it for personalised ads or remarketing. The one use connected with advertising is measuring our own Google ads, below.
If you arrive from one of our Google ads
When you click an AudioRinse ad on Google, the link carries a click id, and our pages then load Google's ad tag. It tells Google about the click and, if you buy Pro after it, about the purchase (its amount, currency and an order number), so we can see which ads lead to customers. It is set not to build remarketing lists or personalise ads. Visitors who don't arrive from an ad never load it, and neither does a browser that sends Global Privacy Control. The tag never sees your recordings, file names or transcripts.
Who processes it for us
- Vercel: hosting the website, anonymous page-view statistics without cookies, and running and storing for the Claude connector and cloud separation (IP address and request logs; pages viewed, referrer, browser, device type and approximate country; recordings sent through Claude and their cleaned versions, deleted within 24 hours; songs sent for cloud separation, deleted when it ends; recordings sent for cloud sheet music, held in memory while the score is written, never stored).
- Replicate: separating a song's vocals on a GPU when you choose Cloud on a vocal remover page, and repairing distorted speech when you press the cloud repair on the de-clipper page (Pro) (a copy of the song's or the recording's sound and the results, deleted within an hour).
- Neon: our database (account email, plan, monthly export counts, ratings, and your starred and recently used tools, each tool's settings, presets and chains, what you use AudioRinse for, your votes and suggestions for the next tools, whether you asked for email about new tools, and your invites).
- Clerk: sign-in and accounts (email address, sign-in method, session cookies).
- Stripe: payments (we never see your card) (name, email, payment details, billing country).
- Google (Gemini API): writing a song's lyrics when you choose Cloud on the song-to-lyrics page; under Google's paid Gemini API terms it isn't used to improve Google's products, and Google keeps requests for a limited time to detect abuse (a mono copy of the song you send, and the lyrics it writes).
- Hugging Face: serving the transcriber's speech and speaker models, the vocal remover's separation models and the sheet-music tool's models, downloaded once by your browser (the recording, the song, the transcript, the stems and the score never leave your device) (IP address and which public model file was downloaded).
- Resend: sending account email (the reminder before a free trial ends) and, if you ask for it, email about new tools (email address and the email's contents).
- Upstash: rate limiting, counting how often each tool is used and starred, and the Claude connector's upload links and jobs (IP address, kept for at most 24 hours; for the tool counts, the tool and the day, and a one-way code made from the IP address and browser, kept 2 days; upload links, job records and the list of your jobs, kept about a day; cloud sheet music's scores (the notes, no sound), kept an hour).
- Google (Google Ads): measuring our Google ads, only if you arrive by clicking one: Google's tag records the click and, if you buy Pro, the purchase, so we know which ads lead to customers. Not used for remarketing or personalised ads, and visitors who don't come from an ad never load it (the ad click's id, the page you land on and the page you return to after checkout, IP address, browser; for a purchase, its amount, currency and an order number (never your name, email or payment details)).
- PostHog: anonymous usage statistics (which tools are used, whether cleaning and downloads succeed, whether a free trial or subscription starts, converts or is cancelled), without cookies (pages viewed, tool events, what you type into the tool search, page load speed, browser type, approximate country; for a subscription, its plan, its status changes and the page it started from, under an id that means nothing outside our payment provider (never your name, email or payment details)).
These providers may process data in the United States. They act on our instructions and under their own security commitments.
Cookies
We use only cookies needed to run the service: the device identifier above, sign-in session cookies from Clerk if you have an account, and Stripe's cookies on its checkout page. The one exception is for visitors who arrive from one of our Google ads (above): Google's tag sets its own cookies on our site (such as _gcl_au and _gcl_aw, kept up to 90 days) to link a later purchase to the ad. In the EEA, the UK and Switzerland it sets none and measures without cookies.
How long we keep it
Account and billing records are kept while your account exists and as long as tax and accounting law require for payments. Monthly export counts are kept for up to 13 months. Ratings and cancellation reasons are kept while they help us improve the tools. Your tools, settings and votes are kept until you change them or delete your account. Deleting your account deletes your account record, export counts, tools, settings and votes, and unlinks your ratings and cancellation reason from you.
Your rights
You can ask to see, correct, export or delete your personal information, or object to how we use it, by emailing privacy@wikivec.com. You can delete your account yourself from the account menu. Depending on where you live (for example the EU, UK or California) you may have further rights and can complain to your local data protection authority.
Children
AudioRinse is not directed at children under 13 (16 in the EEA), and we don't knowingly collect their personal information.
Changes
If we change this policy we'll update this page and its effective date, and tell account holders about material changes by email.
Contact
Wikivec Inc.1111B S Governors Ave STE 7867
Dover, DE 19904
United States
privacy@wikivec.com