Sotto Voce · Privacy Policy

Privacy Policy

Sotto Voce: Private Dictation, for iPhone. Sotto Voce was previously called AudioToText; this policy covers the same app.

The short version

Everything you say or type in Sotto Voce stays on your iPhone. Transcription and AI summarization both run entirely on-device. There are no accounts, no servers of ours, and no analytics — Sotto Voce has nothing to upload, because nothing it does leaves your phone.

The only network request the app makes is downloading the files of an on-device model — a one-way download of model weights, not your data. See Model downloads below.

How transcription and summarization work

Sotto Voce transcribes speech using either Apple’s built-in, on-device speech recognition (“Standard”), or an optional NVIDIA Parakeet model running locally through Apple’s Core ML (“Enhanced”) — whichever you choose on the Model screen. Both run entirely on your device; neither sends audio or text anywhere. Background-noise filtering and separating the speakers in a recording also run on your device.

Meeting notes and summaries are generated entirely on-device using Apple’s Foundation Models framework. No transcript, recording, or summary is ever sent to an external service, including Apple’s own servers.

Why the keyboard needs Full Access

iOS keyboard extensions cannot use the microphone directly. Sotto Voce’s keyboard works around that by having the main app claim the microphone and hand transcribed text to the keyboard through a small piece of storage shared between them (an “App Group”) — and iOS only allows a keyboard extension to read that shared storage once you grant Full Access.

Full Access is used solely for that local, on-device handoff: letting the keyboard know a dictation session is running, and reading back the words it produced. The keyboard makes no network requests of any kind, and Sotto Voce does not use Full Access for analytics or any purpose beyond this.

Model downloads

Apple’s built-in speech recognition ships with iOS and needs no download. Some features instead use an open model that runs on your device — for example the optional Enhanced transcription model, the background-noise filter, and speaker separation for recordings. The first time you use one of these, the app downloads that model’s files from Hugging Face’s servers over the network.

This is a one-way download of files. No audio, transcript, typed text or personal information is uploaded as part of it, and nothing is sent back. Once a model is downloaded it works offline.

What Sotto Voce stores

The app saves what it needs to do its job, and nothing else:

All of it is written to a private container on your iPhone. It is not synced to iCloud, not backed up to any server of ours, and never transmitted. Nobody — including the developer — can see it.

You can delete any recording from within the app. Deleting the app removes everything at once. There is no copy anywhere else to request the deletion of.

Purchases

Sotto Voce Pro is bought through the App Store, and Apple processes the payment under its own terms and privacy policy. Sotto Voce never sees your payment details, and there is no Sotto Voce account to link a purchase to.

What we don’t do

Children

Sotto Voce is a general-audience app and is not directed at children under 13. Because the app collects no data from any user, it does not knowingly collect data from children either.

Your rights

Under UK and EU data protection law you have rights of access, correction, erasure and objection over personal data held about you. For Sotto Voce, those rights are satisfied directly on your device: everything the app stores is visible in the app, can be edited or deleted there, and is destroyed when you delete the app. The developer holds no copy and so has nothing to disclose or erase.

Changes to this policy

If this policy changes it will be updated on this page and the date below will change with it. If a change ever materially affects how your information is handled, it will be described here rather than quietly amended.

Contact

Questions about privacy, or anything else: daniel.li.work+support@gmail.com

Last updated 26 September 2026