Cloud dictation

Cloud dictation is dictation in which speech recognition runs on a remote server instead of on your device. Your microphone records the sound, the audio travels over the internet, and the server returns the text. Because the audio leaves your device, cloud dictation depends on a connection and on the provider's rules for keeping your data.

A common example is speech recognition in a web browser. By default, speech recognition on a web page sends your audio to a server-based recognition service, so it does not work offline ¹. Browser support for on-device recognition differs between browsers and versions, so test the one you use.

Recognition moved to servers because the models were too large for phones

Early recognizers were large. Google said its traditional server-side recognizer used a search graph of almost 2 GB, too large for a phone, which is why it needed a connection ². Later designs changed that. In 2019 Google said its new on-device recognizer design offered the same accuracy as its traditional server-based models in a much smaller model ³.

That claim comes from Google's own tests, for American English in 2019. It does not mean every cloud or on-device tool performs alike, because accuracy still varies by language and device.

Privacy depends on what the provider keeps

Sending audio to a server raises one question: what does the provider keep? Policies differ. Aqua Voice says that for users with Privacy Mode turned off, it may store transcript data on its servers to improve the product ⁴. The section covers transcript data only, so read the full policy before you dictate confidential material.

Compare two setups, not two brand names. A browser tool that calls a server sends each recording out of the device. A desktop tool that runs its recognition model locally keeps the audio on the computer, and it keeps working when the connection drops. Check which setting your app uses, because some apps offer both.

Cloud and on-device dictation trade off speed, offline use and control

Cloud recognition can use larger models than a phone can hold, but it needs a connection and sends your words to a third party. On-device recognition works offline and keeps audio local, but the model has to be downloaded and stored. Neither is always better. Choose by what you dictate: a shopping list is low risk, while client records or medical notes are not.

If you need to know whether dictation works without a connection, see the question on dictation without internet. For the privacy side, read whether voice-to-text is safe and voice data retention. For the technical side, see on-device speech recognition.

Cloud dictation is one kind of dictation software. Many cloud tools also add AI dictation features, which send the text to a language model after recognition. Check each step separately, because the audio and the text may go to different places. If a tool has a free plan, check its limits too, as explained in the entry on free tiers.