If you've been reading about modern voice typing and quietly wondering whether your seven-year-old laptop is going to chug, this post is for you.
There is a popular assumption that anything described as "AI-powered" needs a roaring new gaming PC with a chunky graphics card to run. For some kinds of AI, that is fair enough. For voice typing, it is not. OpenAI's Whisper model, which is the engine behind most of the better modern dictation tools, was specifically designed to run on the kind of normal everyday computers most of us actually own. With the right model size and a sensible configuration, you can get genuinely smooth, accurate dictation on a laptop that's been sitting in your kitchen since 2018.
This is the no-nonsense guide to making that work. We will go through realistic voice typing system requirements, look at how each Whisper model size behaves on different hardware, and cover the practical tweaks that turn a slow setup into a smooth one. If you want to run Whisper on an old laptop, find dictation software for a low-end PC, or just check whether your current machine can handle it before you commit, read on.
The short version: PeekoType runs on Windows 10 and Windows 11, on x64 processors, with no GPU required. 8 GB of RAM and any Intel or AMD CPU from the last seven or eight years will get you a comfortable experience with the small or medium Whisper model. There is a 14-day free trial, so you can confirm performance on your own machine before paying anything.
The basic system requirements
Let's start with the hard floor. The minimum specifications for PeekoType are deliberately modest, because the whole point of an offline tool is to be accessible to people who don't have brand-new hardware.
| Component | Minimum | Comfortable |
|---|---|---|
| Operating system | Windows 10 (64-bit) | Windows 10 or 11 (64-bit) |
| Processor | Any 64-bit Intel or AMD CPU from 2016 onwards | Intel Core i5 / Ryzen 5 (2018 or newer) |
| RAM | 4 GB | 8 GB or more |
| Disk space | 2 GB for app + smallest model | 5 GB if you want larger models too |
| Microphone | Any working mic (built-in is fine) | USB headset mic for best accuracy |
| GPU | Not required | Optional speed boost if you have one |
| Internet | Required only for the initial download | Not required day-to-day |
If your laptop meets the minimum row, you can run the software. If it meets the comfortable row, you will barely notice the dictation happening. Either way, the trial is the most honest test, because real-world performance varies a bit more than the spec sheet suggests.
Picking the right Whisper model for your hardware
The single biggest decision you make as a new voice typing user is which model size to use. Whisper comes in several sizes, and PeekoType lets you pick which one you want to run. The bigger the model, the more accurate the transcription, but also the more memory and processing it needs.
Here is the realistic shape of it on a typical Windows laptop.
Tiny model (~75 MB)
The tiny model is genuinely tiny. It will load on essentially anything that runs Windows 10, including a £200 second-hand ThinkPad. The accuracy is fine for everyday English dictation (emails, notes, shopping lists, basic messages), and the response time is almost instant. It is the best starting point if your laptop is more than seven years old or has only 4 GB of RAM.
What it doesn't handle as confidently: unusual proper nouns, technical vocabulary, heavy regional accents and very fast speech. For most everyday users, those limitations matter less than they sound.
Small model (~250 MB)
The small model is the sweet spot for older PCs. It handles general English dictation beautifully, most accents respectably, and a surprising amount of technical vocabulary. On a laptop with 8 GB of RAM and a mid-range CPU from the last five or six years, the small model gives you near-instant dictation with negligible CPU spikes.
If you are unsure where to start, this is the model to try first. It is the right answer for the majority of users on the majority of older hardware.
Medium model (~1.5 GB)
The medium model is a noticeable step up in accuracy, particularly for unusual words, names, dates, and technical terms. It works comfortably on most laptops from the last three or four years with 8 GB of RAM or more. On a five-year-old machine, it still works, but you may notice a brief pause while the model is loaded into memory.
This is the model most professional writers, researchers and clinicians end up using, because the extra accuracy is worth the slightly higher footprint.
Large model (~3 GB)
The large model is the highest-accuracy option. It is genuinely brilliant for very precise dictation, rare proper nouns, specialist vocabulary and non-English languages. It also needs more memory and a faster CPU, so it is best suited to newer laptops or to desktop machines.
If you have a recent gaming PC, a current-generation laptop with 16 GB of RAM, or anything with a modern GPU, the large model is the one to choose. On older hardware, the marginal accuracy gain is rarely worth the extra resource cost.
Real-world performance on different generations of hardware
Specs are one thing, but most users want to know what it actually feels like. Here are some honest, observed performance notes from users on different generations of hardware. Times below describe the rough lag between you finishing a sentence and the text appearing.
A 2017 ultrabook (Intel Core i5-7200U, 8 GB RAM, no GPU)
This is the kind of laptop a lot of people are still using. Voice typing with the small model feels instant. The text appears as fast as you can speak. The medium model takes about half a second to a second per sentence, which is noticeable but not annoying. The large model is not really recommended on this generation.
A 2019 mid-range laptop (Intel Core i5-10210U, 8 GB RAM, no GPU)
The small and medium models both feel essentially instant. The large model takes about a second per sentence and works fine for most use cases. This is roughly the sweet spot of hardware for offline voice typing.
A 2014 desktop (Intel Core i7-4770, 16 GB RAM, mid-range GPU)
Even on a desktop CPU from over a decade ago, the small and medium models run comfortably. With the older GPU enabled, the medium model is noticeably faster than CPU-only. The large model is workable but a touch slow.
A 2022 modern laptop (Intel Core i7-1255U, 16 GB RAM, no discrete GPU)
Everything is instant. You can run the large model and still feel like you are typing at the speed of thought. This is what brand-new dictation feels like, and it is also the experience that anyone with a recent machine should expect.
A budget Windows laptop (Celeron N4020, 4 GB RAM)
The honest answer for very low-end machines like Celeron and Pentium chips with 4 GB of RAM is that you should stick to the tiny model. Performance is acceptable but not snappy, and the small model will be noticeably slower. You can still use the software productively, but the experience is best described as patient rather than instant.
Why this works without a GPU
People sometimes ask why Whisper-based voice typing doesn't need the kind of expensive graphics card that other AI workloads do. The honest answer is that voice typing is, computationally speaking, a relatively small task. The Whisper model itself is small (a few hundred megabytes at the small size), and the audio it has to process is short (a sentence or two at a time). Modern CPUs handle that comfortably without needing to call on a GPU.
If you do have a recent GPU, PeekoType can take advantage of it for an extra speed boost, but it is genuinely optional. The default CPU-only mode is the route most users go, and it is fast enough that no one feels they are missing out. For the technical specification of how Whisper itself works, OpenAI's original Whisper research page is the place to start.
Tips for getting smoother performance
If your initial run feels a little sluggish, there are a few practical tweaks that often make a noticeable difference.
- Close memory-hungry browser tabs. Modern browsers (especially Chrome) can quietly consume 4 GB of RAM all on their own. If you are on a 4 GB or 8 GB machine, closing thirty tabs before dictating gives the Whisper model the room it needs.
- Drop to a smaller model. If the medium model feels slow, the small model is right there and only slightly less accurate. You can swap models in the PeekoType settings without restarting.
- Disable Windows visual effects. Windows animations and transparency effects use CPU. Turning them off (Settings, Accessibility, Visual effects) is a small free win on older laptops.
- Don't run voice typing during a Windows Update. Background updates and antivirus scans can eat up an extraordinary amount of CPU. If voice typing feels sluggish for no obvious reason, check what Windows is doing in the background first.
- Use a wired headset microphone. Built-in laptop microphones are fine, but they pick up keyboard noise and room reverb, which the model has to work harder to filter out. A basic £20 USB headset noticeably improves both speed and accuracy.
- Keep your laptop plugged in. Windows aggressively throttles CPU performance on battery power, especially on older laptops. Plugged in, your CPU runs at full speed, and dictation feels much snappier.
Why offline really matters for older PCs
There is a specific, practical reason that on-device voice typing is often better suited to older hardware than cloud-based alternatives, even though cloud tools nominally do their heavy lifting on a remote server.
Cloud dictation depends on a constantly stable network connection, microphone audio streaming in real time over your home broadband, and your laptop's ability to keep that connection alive while also running everything else you are doing. On an older laptop with a slightly flaky WiFi card or limited memory, that whole pipeline tends to be the weakest link. The transcription itself happens on a fast server, but the experience is gated by your network and your local resources.
Offline voice typing avoids the entire issue. Your audio doesn't leave your machine, the model runs locally, and the network is no longer in the loop. On older hardware in particular, this often turns out to be more responsive than cloud dictation, not less. You also get the privacy benefit, which we cover in our GDPR-by-design article.
What about Windows 11 ARM, Surface devices and Snapdragon laptops?
The current generation of ARM-based Windows laptops (the Snapdragon X Elite Surface devices, for example) is supported in compatibility mode. Performance is reasonable, particularly for the small and medium models, although native ARM support is on the development roadmap and will improve performance further. If you are on a Copilot+ PC or a recent ARM laptop, the trial is the right way to confirm the experience on your specific machine.
Is it worth running this on a really old machine?
Honest answer. If your laptop is from before 2014, or runs only Windows 7 or 8, or has 2 GB of RAM, then voice typing is going to be a struggle regardless of which tool you choose. Whisper itself simply isn't a viable workload for hardware of that vintage, and neither is any of the modern alternatives.
If, however, your laptop is from roughly 2016 onwards, runs Windows 10 or 11, and has 4 GB of RAM or more, you are well inside the range where offline voice typing works comfortably. The free trial is the easiest way to find out for sure. Install it, pick the small model, dictate a paragraph, and see how it feels.
How this compares with the alternatives on old hardware
One useful frame of reference is how the older options compare on the same older laptop.
- Dragon NaturallySpeaking has historically been quite resource-hungry, particularly on older machines, and requires noticeable installation time and ongoing voice training. Our Dragon dictation alternative guide goes into the comparison in detail.
- Windows Speech Recognition (the built-in tool) is lightweight, but the accuracy is dated and the workflow is fiddly. Our guide to hands-free typing on Windows covers it as one of the options.
- Cloud dictation tools (Wispr Flow, Otter, Google Voice Typing) shift the compute load to the cloud, but they require a steady internet connection and you get whatever the network is giving you on the day.
- PeekoType sits in the sweet spot of being modern (Whisper-based, high accuracy), light (no GPU required, small CPU footprint at the smaller model sizes), private (offline) and affordable (£19.99 once).
If you are weighing up options across the whole market, our honest review of every voice-to-text app covers the field in more detail.
The one-page checklist for older PCs
If you have skimmed this far and just want the short version, here it is.
- Confirm you are on Windows 10 or 11, 64-bit.
- Confirm you have at least 4 GB of RAM and any CPU from roughly 2016 onwards.
- Download the free trial of PeekoType. No card required.
- On the first run, pick the small Whisper model.
- Plug your laptop into mains power, close any unused browser tabs, and try a paragraph of dictation.
- If it feels great, you are done. If it feels slow, drop to the tiny model. If it feels too slow even then, your hardware is probably below the practical floor for this kind of tool.
For most users, step five is where the journey ends. The dictation works, the laptop is happier than expected, and the new habit settles in. Our setup guide for beginners covers the workflow side once the hardware question is settled.
The bottom line
You almost certainly do not need a new laptop to do voice typing well in 2026. The hardware demands of modern offline dictation are much more modest than the marketing for AI tools would suggest, and the model size options give you a way to match the workload to whatever computer you happen to be on. If your laptop has been quietly waiting in a drawer because you thought it was too slow for "AI stuff", get it out, give it a clean, and try the free trial. You may be pleasantly surprised.
If you'd like to confirm whether your specific machine will work well, please drop us an email at support@peekotype.com with a rough description of your hardware. We will give you an honest answer rather than nudging you towards a setup that doesn't suit you. The whole point of an offline tool is that it should work on the computer you already have.