Does dictation software send your voice to the cloud?

What actually leaves your computer when you dictate: the three kinds of dictation tool, how to check which one you have, and what it means for work under an NDA.

7 min read

In short

  • It depends on the tool. Cloud dictation uploads your audio to a server, on-device dictation transcribes on your own computer, and hybrid tools split the job.
  • You can check for yourself: switch off the internet and dictate, read the privacy policy for the words retention and training, and watch the app's network activity.
  • Local dictation keeps your voice on your machine. The assistant you paste the words into still receives them, so its rules apply as before.
  • Dictate Safe is local by default, shows which of three states it is in, and has one switch that blocks every outbound connection.

What happens when you press the dictation key?

Every dictation tool does the same 4 things, in this order:

  1. Your microphone records audio.
  2. Speech recognition turns the audio into text.
  3. Optionally, a cleanup step removes filler words, fixes punctuation or tidies the wording.
  4. The text is typed into the app you are using.

Steps 1 and 4 always happen on your computer. Steps 2 and 3 decide what leaves it, because each runs either on your machine or on someone else’s server.

What are the 3 kinds of dictation tool?

KindSpeech recognitionCleanupLeaves your computerWorks with no internet
CloudOn a serverOn a serverThe audioNo
On-deviceOn your computerOn your computer, or noneNothingYes
HybridOn your computerOn a serverThe text, not the audioThe speech yes, the cleanup no

Cloud tools can run very large models and need no setup. On-device tools give that up to keep everything on the machine. Whichever you choose, find out whether a setting can change where the audio goes, and whether the app tells you when it does.

A tool that transcribes on a server may keep the recording or delete it, and policies change, so read the current policy.

How can you check what a tool really does?

Five checks need no technical skill and no help from the vendor:

  • Turn off the internet and dictate. If it still works at normal speed, speech recognition is happening on your computer. If it fails or stalls, it needs a server.
  • Read the privacy policy for 4 words: audio, retention, training and sub-processors. You want to know whether recordings are kept, for how long, whether they are used to improve models, and which other companies handle them.
  • Watch the network. On a Mac, the Network tab of Activity Monitor shows how much each program has sent. On Windows, the Network tab of Resource Monitor does the same. A firewall that reports outgoing connections shows where they go.
  • Find out where your history lives. A tool that saves what you dictated should say where, and let you delete both the text and the recording.
  • Notice whether it needs an account. A tool that works without one holds less about you, although an account alone does not mean audio is uploaded.

Why does it matter at work?

Many NDAs and company AI policies are written about where information is allowed to go. A recording of you reading out a client’s draft contract is information, and once it is uploaded to a vendor’s server it is outside your control, however carefully the vendor handles it. In regulated work such as law, finance or health research, the question is usually asked before a tool is approved.

On-device dictation answers that question simply, because there is nothing uploaded to assess. Dictate Safe has no compliance certification. By default, nothing leaves the computer.

What does local dictation not cover?

  • The assistant you paste into. If you dictate a prompt and paste it into a chatbot, those words go to that chatbot. Where your voice was processed and where your words end up are two separate questions.
  • Files on your own computer. Local means the recordings and history sit in your user account on your machine. Decide how long you keep them, and delete what you no longer need.
  • Optional cloud features. Many local tools also offer a cloud option for speed. Find out exactly what switching it on sends: the audio, the text, or both.

How does Dictate Safe handle it?

This is our own product, Dictate Safe, a dictation app, so read it as a description and not a neutral review. The app is in one of three states at any time, and it shows a badge for the current one on every screen.

StateWhat leaves your computer
Local (the default)Nothing.
Hybrid: local speech, cloud cleanupThe transcript text only, never the audio, sent to a provider you chose.
Cloud: cloud speechYour recording, sent to Groq with your own key.

The cloud options use your own key: Groq for cloud speech, Groq or OpenAI for cloud cleanup. The app talks to the provider directly, so we never see the audio, the text or the key. There is no telemetry, no analytics and no tracking.

Apart from a cloud provider you switch on, the app makes only a few connections: to download a speech model, to sign in, to check for a new version, and to download the prompt library when you ask for it. Offline Lock is a single switch that refuses every one of them, ours included, and says so in plain words when it does.

You can also delete any dictation, singly or in bulk. The text and the recording are erased from your computer, and the freed space is overwritten.

More on this in the security overview and the help article on where your data is stored.

Questions people ask

Does dictation software send my voice to the cloud?

Cloud dictation tools do: the audio is uploaded to a server and transcribed there. On-device tools transcribe on your own computer and upload nothing. Hybrid tools transcribe locally and send only the text to a server for cleanup.

Do I need an internet connection to dictate?

Not with an on-device tool once it is set up. Dictate Safe needs the internet once, to download a speech model, and works offline afterwards.

Is on-device dictation less accurate than cloud dictation?

Not necessarily, and it varies from tool to tool. Results depend a great deal on your voice, your microphone and the words you use, and a vocabulary list helps with names and jargon in any tool that has one. The fair check is your own: dictate the same paragraph, full of the terms you really use, into each tool you are considering and compare the text.

Does Dictate Safe collect telemetry or analytics?

No. The app has no telemetry, no analytics and no tracking, and we cannot see your dictations.

What does Dictate Safe connect to?

It connects to download speech models, to sign in and check for updates, to download the prompt library when you ask for it, and to a cloud provider only if you switch one on with your own key. Offline Lock blocks all of those with one switch.

Can I delete my dictation history?

Yes. You can delete any dictation singly or in bulk. The text and the recording are erased from your computer, and the freed space is overwritten.

Can I use it under an NDA or a company policy?

Dictate Safe is fully local by default, so your recordings and transcripts are not sent anywhere. Whether a tool is allowed under a particular NDA or policy is for you or your employer to decide, and the assistant you paste your words into is a separate question.

Keep reading