Privacy Policy

What we collect

When you create an account, we store your email address and a display name. When you use AudLit, we store project metadata (book titles, chapter names, section structure), the parsed paragraph text of any manuscript you import so we can render it in the teleprompter, and any annotations or comments created during collaboration. Each time you bring in a manuscript we note its file type (DOCX, PDF, EPUB and so on, or that you pasted the text) and whether AudLit could read it, never the file’s name. We also store optional content you choose to add: custom vocabulary entries, character notes, narrator voice-library entries, and audition scripts.

AI and your voice — our human-first commitment

AudLit is a production tool for human narrators. It does not build, train, sell, or offer AI-generated, synthetic, or cloned voices — there is no voice-generation or voice-cloning feature in AudLit.

• No training. We do not use your recordings, your voice, or your manuscripts to create or train any AI, machine-learning, or synthetic-voice model — ours or anyone else's. Our transcription provider (AssemblyAI) receives audio solely to convert speech to text, and that audio is deleted after processing. Our text provider (Anthropic) does not train models on API inputs or outputs.
• No synthetic voices. AudLit never creates a synthetic or cloned copy of your voice. Every recording in AudLit is your own performance.
• Consent. Because we create no synthetic voice models, we make no use of your voice, image, name, or likeness for that purpose — and never would without your active, informed, written consent.
• Control. Your recordings and manuscripts are yours. Raw recordings are stored in your own local folder on your device; you can play, export, edit, and delete your work at any time.
• Compensation. We never use your voice or files to build or monetize voice models. Any future use beyond operating the service for you would only occur under a separate, mutually agreed arrangement with fair compensation.
• Security. We make good-faith efforts to protect your files from unauthorized third-party access: audio stays local where possible, uploaded transcription audio is deleted promptly, data access is scoped to your account, and download links are short-lived.
• Sourcing. AudLit does not build, train, or operate any synthetic-voice or voice-cloning model, so we do not source voice data of any kind: not from narrators, not from third parties, and not by scraping or licensing recordings. Your recordings are used only to provide the service to you. The third-party AI services we use are listed on our subprocessors page, and we give none of them your content to train on.

Local-first architecture

The takes you record are saved directly to a folder you choose on your own device, and our servers never read or store those original files.

One copy of your audio is stored by us on an ongoing basis: when you save or master a chapter, the mastered MP3 is uploaded to our storage under your account so it can be played on the review page you share, and it stays there until you delete the chapter or project. Audio sent for transcription is covered separately below.

We never use your recordings to train any AI model, and we never create a synthetic or cloned copy of your voice.

Follow mode. The teleprompter's follow mode keeps the script moving at your reading pace by listening to you. By default it listens through the microphone you actually record with and streams that audio to AssemblyAI, under our contract with them. They retain no follow-mode audio or transcripts at all, and our opt-out from model training, benchmarking and deidentified training covers both AssemblyAI and the infrastructure providers they run on. Their own subprocessor list is published in their trust center, linked from our subprocessors page. Follow mode runs only while you are recording. Timed scrolling and manual scrolling do not listen at all, and you can switch to them at any time.

When follow mode uses your browser instead. AssemblyAI's live service covers English, Spanish, German, French, Italian and Portuguese. For a book in another language, or if you choose it yourself, follow mode uses your browser's own speech recognition service instead, which on Chrome and Edge is Google's. Your browser then streams microphone audio to Google directly: AudLit never receives or stores that audio, and we use only the recognized words to work out your position on the page. Because this is a browser feature rather than a service we contract for, that audio is handled under Google's terms rather than ours. The notice you see before follow mode records names both routes and lets you choose between them.

An earlier version of our in-app notice said all follow-mode processing was local. That was accurate when written, before follow mode used browser speech recognition, and we did not update it when the feature changed. We have corrected it and now ask everyone to read the accurate description before follow mode records.

Your original manuscript file also stays on your computer. When you import a manuscript, AudLit opens it directly from your local file system using the File System Access API, extracts the paragraph text in your browser, and sends only the extracted text to our servers for structure detection. The original DOCX, PDF, EPUB, or other manuscript file is never uploaded to AudLit, never sent to our AI provider, and never stored on any AudLit server.

The parsed paragraph text of your manuscript is stored server-side so the teleprompter, cleanup pipeline, and review page can function. This text is the minimum data we need to run AudLit's core features.

Audio uploaded for transcription

When you analyze a take or run cleanup, the audio file for the section is uploaded to our transcription provider for speech-to-text processing. We delete the uploaded file from our servers as soon as transcription completes, and a daily cleanup job sweeps any uploads that were not deleted inline. Our transcription provider is contractually required to delete uploaded audio and transcripts on their side; the current provider is listed on our subprocessor page.

We keep the resulting transcript text and word-level timestamps in our database so that running cleanup, opening a previously analyzed take, or generating review timestamps does not require re-uploading the audio. Cached transcripts are deleted when you delete the associated project.

AI features that process manuscript content

Several features send portions of your manuscript or transcript text to a third-party AI provider so the model can perform a specific task. Each call sends only the content needed for that feature:

• Chapter and heading detection — invoked when you import a manuscript. Sends a sampled representation of your manuscript paragraphs so the model can identify chapter boundaries and headings.
• PDF title cleanup — invoked on PDF imports as a second pass. Sends detected heading candidates with short surrounding-paragraph context so the model can repair garbled titles produced by PDF text extraction.
• Character detection — invoked when you ask AudLit to auto-detect characters. Sends manuscript text so the model can identify named characters, descriptions, and dialogue locations.
• Cleanup-feedback analyzer — invoked after a cleanup run to categorize errors. Sends the manuscript text alongside the raw and cleaned ASR transcripts so the model can distinguish narrator skips from cleanup-pipeline issues.
• Phonetics generation — invoked when you ask AudLit to generate pronunciation guides. Sends the word or character name being phoneticized.
• Accent overlay — invoked when you generate accent-specific respellings on the Character Practice page. Sends the practice scratchpad text you have typed or pasted in (up to ~12,000 characters) plus the accent name and any accent notes you wrote for the character. Your full manuscript is never sent through this feature; only the text in the practice scratchpad is.

Our AI provider does not train models on API inputs or outputs. Inputs and outputs may be retained for up to 30 days for trust and safety review by the AI provider, then deleted. See our subprocessor page for the current provider and a link to their retention terms.

Cleanup-issue reports

AudLit lets you submit feedback when the cleanup pipeline produces a result you believe is incorrect. We designed this feature deliberately so that filing a report does not require sharing manuscript prose with us.

When you submit a cleanup issue, we receive only locator metadata — a manuscript hash, paragraph and word indices, gap timing, your local audio filename, and the description you choose to write. No manuscript text or recorded audio is transmitted through the cleanup-report channel. We replay the issue against the manuscript copy already in your project storage.

We retain submitted reports indefinitely so we can investigate and improve the pipeline. You can withdraw a report at any time from the Submitted Reports panel in the record view; a withdrawn report is removed from your view and our operator queue.

Diagnostic logs

When you run cleanup, AudLit records a per-run diagnostic trace server-side: a manuscript hash, the resolved pipeline settings for that run, the per-paragraph and per-gap decisions the pipeline made, and structural facts about the manuscript (paragraph types, anomaly flags). These records exist so that when you or another user submits a cleanup report, we can reproduce the exact pipeline state that produced the result.

Diagnostic logs contain locator metadata and pipeline state, not manuscript prose or audio. They are retained for 90 days, except the log of a cleanup run you submit a cleanup report about, which we keep so the problem can be reproduced. All of them are deleted when you delete the associated project.

Recording measurements

When you record a take or run the mic setup, AudLit stores measurements about the audio, never the audio itself: levels (peak, loudness and noise floor), whether anything clipped and where in the take, the microphone your browser used (its name and the random identifier your browser assigns it for this site), and your mic setup results: the mic type you chose, how loud your room and voice measured, and what the setup recommended. AudLit also notes which browser and operating system you open it with (their names and versions, whether the device is a phone or a computer, and its processor type, memory class and core count), so that a problem you report can be read against the machine it happened on. Nothing in that record identifies you or your computer.

Personal names attached to your microphone or other equipment are stripped out before anything is stored: “Sam’s AirPods Pro” is stored as “AirPods Pro”. A device renamed to nothing but a name can’t be told apart from a brand name, so if you’ve done that, rename it in your computer’s sound settings.

We use these measurements to help you when something sounds wrong and to improve how AudLit guides you. They contain no recordings and none of your manuscript. They are kept while the project exists and deleted when you delete it.

Third-party services

AudLit uses third-party services for transcription, AI processing, payment, hosting, email, and error reporting. The current providers, their roles, and what is sent to each are listed on our subprocessor page. We maintain that list separately so we can update providers without re-versioning this policy.

Cookies

We use an authentication cookie managed by Supabase to keep you logged in. If you arrive through a partner's referral link, we also set a referral cookie that holds that partner's code and nothing else. It is kept for one year, read only when you create an account and when you buy an Annual plan so we can credit the partner, and never shared with anyone. The partner sees how many people signed up and bought through their link, never who. We do not use analytics cookies, tracking pixels, or any third-party advertising cookies.

Newsletter

AudLit's newsletter is optional. You are only added if you choose it, at signup, when the app asks, or in Settings. While you are subscribed, your email address and first name are held by Mailchimp, which sends it for us; new subscribers confirm by clicking a link in Mailchimp's first email. Every issue has an unsubscribe link, and turning the newsletter off in Settings unsubscribes you too.

No tracking, no ads, no data selling

AudLit does not run analytics or tracking scripts, does not display ads, and does not sell, rent, or share your data with anyone for marketing purposes.

Your California privacy rights (CCPA / CPRA)

If you are a California resident, the California Consumer Privacy Act gives you certain rights regarding personal information we hold about you.

Categories of personal information we collect: identifiers (email, display name, account ID); commercial information (subscription status, hour purchases, billing metadata via Stripe); internet activity (account-scoped application logs and diagnostic traces); user content (manuscript text you import, vocabulary entries, annotations, voice-library entries, submitted feedback). We do not collect geolocation, biometric identifiers, sensory data beyond audio you explicitly upload for transcription, or inferences for profiling.

Sale or sharing: We do not sell or share your personal information as those terms are defined under the CCPA/CPRA.

Your rights: You have the right to know what personal information we hold, the right to request deletion, the right to correct inaccurate information, and the right to limit certain uses of sensitive personal information. To exercise any of these rights, email hugh@audlit.com or use our contact form. We will respond within the timeframes required by law and will not discriminate against you for exercising your rights.

Data deletion

You can delete an individual project from its settings page, or delete your account from your user settings. Deleting a project removes its parsed manuscript paragraphs, transcription cache, annotations, submitted cleanup-issue reports, and diagnostic logs for that project. Deleting your account removes all of the above across every project you own, along with your vocabulary entries, voice-library entries, profile, and billing metadata.

Deletion is purged from active storage within 30 days.

Your audio recordings, kept on your computer via the File System Access API, are unaffected by deletion.

Contact

Questions about this policy? Reach us at hugh@audlit.com or through our contact form.