Free Voice Input: Local Whisper and Ready-Made Apps

In the example described here, almost everything is dictated by voice: prompts for Claude, notes, posts, and replies in messaging apps. Speaking is 3–4 times faster than typing on a keyboard. Local Whisper can recognize the recording on your computer. The destination is a separate privacy boundary: text pasted into a cloud chat or submitted to an online service is then sent according to that service’s behavior.

This guide gives a direct app installation path and a short acceptance test before you rely on dictation. An agent can help investigate a specific failure after that. We also compare the convenience of ready-made paid apps with a local option.

Why dictate by voice instead of typing?#

When you work with AI, the biggest time sink isn't code but wording. You explain to Claude what to do, revise, and clarify. Typing those explanations takes a long time. Speaking them is quick.

The numbers are simple: speaking is about three times faster than typing. That's the difference between “typed a paragraph in two minutes” and “spoke it in thirty seconds.” Over a day, that adds up to tens of minutes saved.

Plus, your hands are free. You can walk around the room, look at another screen, or hold a cup. For many people, it also removes a barrier: speaking is easier than sitting down to put something into writing.

Wispr Flow, superwhisper, Whisper: What's the difference?#

It's easy to get confused by the names, so let's take them in order.

Whisper (without an “r” at the end, from OpenAI) is an engine: an open model that turns speech into text. It's free, open source, and can run directly on your computer. On its own, it's just a “motor” without buttons or windows—ready-made apps use it under the hood.

Wispr Flow (with an “r,” wisprflow.ai) is a ready-made paid app for Mac and Windows. Hold down a key, speak, and the text appears in any field where the cursor is. Convenient, polished, and works out of the box.

superwhisper (superwhisper.com) is a similar app for Mac, Windows, and iPhone. It has a free mode using local models, while the full version costs money. The idea is the same: hold down a key, speak, and get text.

Both approaches turn speech into text, but the model, editing features, and processing location can differ. A ready-made app reduces setup work; local Whisper gives you control of an offline recognition path. Test the same phrase, including names and specialist terms, before deciding which works better for you.

Paid or free: What do Wispr Flow and superwhisper charge for?#

Here's the honest picture of the paid options:

What you actually pay for in paid apps:

• Ready-made interface. Install it and it works; nothing to configure.
• Voice commands. “Comma,” “new paragraph,” and “delete the last thing” are recognized as editing commands (in paid versions).
• Sync and dictionary. Names, terms, and your personal dictionary sync across devices.

All of this is nice. But speech recognition there uses the same class of models available for free. In the local setup described here, no subscription is needed: you speak, and text appears.

Voice input can reduce the friction of drafting long prompts. Install and test one selected app first; if something fails, give an agent the exact app version, selected model, operating system, and observed symptom instead of asking it to invent an unspecified setup.

How does local Whisper work, and why is it private?#

Cloud services work like this: your voice is sent to someone else's server, where it's recognized, and the text is sent back. They charge for this (server rental), and you have to trust the company with your speech.

Local Whisper works differently. You download the model once (a file a couple of hundred megabytes or a little more), and from then on everything happens on your device:

1. Hold down a hotkey and speak

   The microphone records audio while the key is held down.

2. The model on your computer recognizes the speech

   The audio is turned into text right on the device, without the internet.

3. The text is inserted where the cursor is

   Into a chat with Claude, a document, a browser address bar—anywhere.

A local recognition path needs no paid speech endpoint and can transcribe without an internet connection once the model is downloaded. Verify that behavior in a local text editor. It does not stop a cloud chat, clipboard-sync feature, or other destination from transmitting the resulting text later.

Whisper understands about 99 languages, including less common ones like Kazakh and Tatar. The models come in different sizes—from the lightest (fast but rougher) to the largest (more accurate but heavier).

What kind of computer can handle voice input?#

Compatibility and speed depend on the selected app build, model, language, and hardware. Check the installer’s supported operating system and processor architecture before downloading. A larger model needs more memory and storage; a working installation alone does not prove that recognition is fast enough for your workflow.

For the Handy path below, start with a local Whisper Small model and one short phrase. Check the download size shown by the app and wait for it to finish. If loading fails or the delay is too long, inspect the app’s requirements and try a smaller supported local model rather than assuming every older laptop will perform the same. Compare the actual transcript and delay on your computer; no universal hardware minimum or completion time is established by this guide.

One documented local dictation setup for macOS or Windows#

Use Handy v0.9.8 as the concrete app for this recipe. Its versioned README describes local Whisper transcription, configurable shortcuts, and microphone/accessibility permissions. This is a documentation-based recipe; installation and transcription were not executed for this guide.

1. Download the installer for your operating system and processor from that release’s official assets. Install and open Handy. Check that you opened the intended release, rather than a similarly named app from another site.
2. Download and select a local Whisper Small model in the app. Wait until the download has completed. Leave optional online post-processing disabled for the local-recognition check.
3. Allow microphone access for Handy. On macOS, grant its requested Accessibility permission for insertion into other apps in System Settings → Privacy & Security. Review the named app and purpose; do not approve unrelated prompts automatically.
4. In Handy’s Settings, configure the transcription shortcut as Ctrl+Shift+Space and choose a hold-to-record mode. This is a chosen binding, not a promised system default. If the app or another program rejects or owns it, select an available shortcut and use that same shortcut in the checks below.
5. Open a local plain-text document, place the cursor in it, hold the configured shortcut, say “Voice input test, project alpha,” then release. The recording should stop and recognized text should be inserted in that document. Check the words and delay; punctuation may differ.
6. Once the app and model are downloaded, disconnect from the network and repeat the harmless phrase in the local document. A transcript produced there demonstrates that this selected transcription path can work offline. Reconnect afterward. This does not establish that a cloud chat receiving the transcript will keep it local.

If you want an agent’s help adapting this setup, give it a bounded diagnostic prompt:

I am using Handy v0.9.8 with a local Whisper Small model.
My operating system and processor are: [fill in actual values].
My chosen transcription shortcut is: [fill in the configured shortcut].
Observed result of the short local-editor test: [describe what happened].
Check the official instructions for this version. Identify the failing
stage: model download/loading, microphone capture, shortcut, or text insertion.
Explain each proposed permission or change before I apply it. Do not
switch to a cloud transcription API, buy anything, or delete existing files.

Keep the prompt as optional help with a specific observed failure. An agent’s “ready” message and clicking Allow do not replace the transcription and offline checks.

How to check that it works (and what to do if it doesn't)#

After configuring the app, downloaded model, and shortcut, run the local-editor check yourself:

1. Place the cursor in a local text document

   Use a harmless test document for the local-recognition check.

2. Hold the transcription shortcut configured in Handy

   If you selected Ctrl+Shift+Space, use that binding; if you chose another available shortcut, use that one instead.

3. Say a short phrase out loud

   For example: “voice input test.”

4. Release the keys

   The text should appear where the cursor was.

If no text appears, don't worry; these are common issues. Copy this prompt into the Claude chat, and it'll walk you through it:

My configured transcription shortcut in Handy is: [actual binding].
When I hold it, speak, and release it in a local test document,
no text appears.

Check exactly what is broken (no microphone access,
no permission to control the computer, a system hotkey conflict,
or the model wasn't downloaded) and walk me through
fixing it step by step.

Start with these specific stages; other causes are possible:

• No microphone access. The system hasn't allowed the program to hear audio—you need to allow it in the privacy settings.
• No text-insertion permission (Mac). Check Handy’s requested Accessibility permission in System Settings → Privacy & Security, and retry in the local text editor.
• Hotkey conflict. Ctrl+Shift+Space is already used by another program; choose a different combination.
• The model did not finish downloading or loading. Inspect its status in the app and complete the selected local model download before testing again. Do not switch silently to an online transcription service.

What if nothing installs?#

Sometimes there's no time to troubleshoot, but you want to dictate today. In that case, turn on the system's built-in voice input; it's free:

• Mac: “System Settings” → “Keyboard” → “Dictation” → turn it on. On Apple chips, recognition runs offline, right on the device.
• Windows: press Win + H to open the voice typing panel. Keep in mind: it works over the internet (recognition runs on Microsoft's servers), so it isn't private offline processing.

It works right now, with no installation. The quality is more modest than local Whisper. But as a way to avoid getting stuck and start writing by voice today, it's perfectly good.

Wispr Flow or free Whisper: Which should you choose?#

A quick guide to choosing:

The local option avoids a recognition subscription and gives you a path to offline transcription. Confirm its quality, speed, and offline behavior on your computer before relying on it. A commercial app may suit you better if its workflow saves more time; compare using the same harmless phrase.

Just try dictating your first prompt. In the experience described here, after a couple of days of voice input, people almost stopped typing long texts on the keyboard.