A paragraph can look fine on screen and still sound awkward when you read it aloud. Hearing it gives you another way to check a presentation opening, an explanation or a short email before you use it.

OGAD (Off Grid AI Desktop) can read English assistant replies aloud on your Windows PC with a local voice. Install beta 0.0.54-beta.108, prepare an English voice while connected, then listen without internet. You can also generate the reply with a downloaded local chat model, keeping both stages on your computer.

Download OGAD beta for Windows

---

What would you like to do with Off Grid AI?

Have a feature or use case you would like us to support? Tell us what you want to do and which device you use.

Write to support@offgridmobileai.co, join our Slack community, or talk to us on Reddit.

What can you use local speech for?

Start with text you already want to review. A short passage gives you a useful result without preparing a large document or a recording session.

Task	What to listen for
Rehearse a presentation opening	Long sentences, repeated points and awkward transitions
Review an email draft	Whether the request and next action are clear
Hear an explanation	Terms you want the chat model to explain more simply
Check a short checklist	Missing steps or an order that is hard to follow

The voice reads the assistant’s text. It does not check facts, rewrite the message or translate it. Keep the written reply visible so you can compare what you hear with what you intend to say.

What do you need on Windows?

Use the Windows x64 installer from the linked beta release. This guide depends on its packaged local speech runtime; the older stable 0.0.51 build is not the same setup.

For the Windows route, choose an English US or English UK voice. Beta108’s local Windows speech path supports those English voices. The broader multilingual voice options used on Mac are not established as working on Windows in this build. A language appearing in a shared control is not enough to make that language’s runtime available.

You need internet for the app and first-use voice resources. If you also want new AI replies while offline, download a local chat model that fits your PC. Speech playback and chat generation use separate resources. A working text model does not mean the voice files are ready.

Local speech is a core feature. You do not need Pro meeting recording or an optional NVIDIA performance pack to use its CPU route. You also do not need to enable your microphone to listen to text.

How do you set up an English voice?

1. Install and open the beta Windows app.
2. In Models → Voice, prepare the local voice model and select Use. If you previously chose a remote speech provider, switch back to the local model for this guide.
3. In Chat, open the Settings button to open Model settings, then choose Voice.
4. Set Interface mode to Chat, then turn Text-to-speech on. This shows playback actions on assistant messages.
5. Under Language, choose English US or English UK. Then choose a Voice in that language.
6. Wait while OGAD checks or downloads its files. Continue when the status says the voice is ready.
7. Select Test voice. Check that you hear the sample through your intended headphones or speakers.

Changing the selected voice can require additional preparation. Finish that step while connected. Leave playback speed at its default for your first check so you can assess the voice clearly.

The released voice settings prepare the selected voice and show its download state. You do not need to install a separate cloud speech service or supply its API key for this local route.

How do you hear your first useful paragraph?

Select a downloaded local text model and open Chat. Give it a small task with a clear purpose:

Write a 70-word opening for a presentation about starting a community garden. Explain the idea and invite volunteers. Do not invent a location, date or attendance figure.

Read the completed reply. Check that it follows the brief, then use Speak on that assistant message.

Listen for one thing you want to improve. If the opening takes too long to reach the request, ask for a shorter version that puts the invitation first. Review the revision, play it and keep the version you prefer. The useful outcome is a paragraph you have checked both on screen and aloud.

To hear your own draft, ask the local chat model to repeat a short passage exactly. Compare its response with the original before using Speak: a chat model can change words even when asked to copy them. This workflow reads assistant messages inside OGAD; it is not a system-wide Windows screen reader.

How can you check that it works offline?

Finish the downloads and play one short reply while connected. Then disconnect Wi-Fi and Ethernet.

Generate a new short reply with the local chat model and select Speak. This checks a new local synthesis request, rather than only replaying audio that was already saved. If you only want to check speech, use a completed reply that you have not played before.

If chat works but speech asks for files, reconnect and finish preparing the selected voice. If speech works but a new answer fails, check the selected chat model separately. Keep remote providers and web requests out of this offline check.

What if speech is missing or stops early?

Problem	Check and next action
There is no Speak action	Use Chat mode, enable Text-to-speech and wait for the reply to complete
Voice setup reports an error	Check your connection, use an English US or UK voice, and select Retry
Playback is silent	Check Windows output device and volume, then try Test voice
A voice works online but fails offline	Complete that selected voice’s resource download before disconnecting
A long passage is cut short	Split it into shorter replies and play each one

The released speech service limits each synthesis call to 2,000 characters. Start with a short paragraph; do not treat this as a one-click narrator for an entire book. Check names, abbreviations and numbers by listening to the actual result.

Install OGAD beta for Windows, prepare one English voice and listen to a paragraph you want to improve. Once its files are ready, you can keep that review step on your own PC without an internet connection.