Does on-device AI dictation drain your battery?

Yes, on-device dictation uses battery, because your computer does the speech work itself rather than sending the audio to a server. How much it uses is not published for laptops or phones, as far as we could find. So we cannot give you a reliable battery-hours figure for dictation today. What we can say is what drives the energy use, and how to measure it on your own Mac.

The figures on this page were checked on October 17, 2026. Energy use depends on the hardware and on the app's version, so a result from one machine does not carry over to another.

We found no published test of dictation on a laptop or phone

We searched for measured battery figures for on-device speech recognition on laptops and phones, and we found no test of a dictation session. The research we found measures other things, such as a streaming model's energy use. The closest rigorous result concerns a streaming model. Li and colleagues report that their component-level design changes cut the energy use of an on-device streaming speech recognition model by up to 47 percent while keeping accuracy comparable ¹. That is a best case for one model design, measured by its authors, and it is not a battery-life figure for a dictation app.

Until a measurement like that exists for dictation on a consumer device, any battery-hours figure you see should be treated with suspicion. Ask how it was measured, on which machine, and with which model.

The model and the chip set most of the energy

Energy use depends on what runs and where it runs. Apple describes the Apple Neural Engine as the energy-efficient and high-throughput engine for machine learning inference on Apple silicon ². Apple presents that engine as its efficient path for this kind of work, so the chip and the code path both matter. An app that uses the Neural Engine may use less energy than one that does not, but we did not find a published comparison of dictation apps to confirm the size of that difference.

The size of the model matters as well. The large Whisper model in whisper.cpp needs about 3.9 GB of memory ³. Quantized models need less memory and disk space, and can run more efficiently on some hardware ⁴. A smaller model needs less memory, but whether it uses less energy depends on the hardware, and we could not measure that for a dictation session.

Continuous listening costs more than occasional dictation

The Li study shows that component-level design changes can cut a streaming model's energy use ¹. The next point is our reasoning, not a measurement: a model that runs continuously, as a streaming recognizer does, uses energy for as long as it listens. A dictation app that you start for a few sentences and then stop does less work overall. Long sessions and always-on listening are the cases where the battery cost is most likely to show.

You can measure the cost yourself in Activity Monitor on a Mac

The Energy pane of Activity Monitor shows the energy use of each app and of the whole computer. The column that matters here is Energy Impact, which Apple describes as a relative measure of the current energy consumption of the app, where lower is better ⁵. Apple gives it no units, so it can rank one run against another but cannot give you watt-hours or battery minutes.

To compare the cost of dictation on your own Mac:

1. Open Activity Monitor from Applications, then Utilities, and choose the Energy pane.
2. Close the dictation app, wait a minute, and note the Energy Impact of the whole system.
3. Open the dictation app and dictate for five minutes at your normal pace, then read its Energy Impact and note how it changes.
4. Repeat the test on the same Mac, the same day and the same microphone, so that the runs are comparable.

Column names and availability vary by macOS version, so use the names that appear on your screen.

It depends on how you dictate and on which device

The cost is not the same for every user. The table below sets out the factors that change it.

What to do if battery life matters to you

1. Run the Activity Monitor test above on the machine you actually use, with the model you actually use.
2. Choose the smallest model that meets your accuracy needs, if the app lets you choose one.
3. Plug in for long sessions. A long dictation session on battery is likely to drain it faster than a few sentences.
4. Watch for steady use. If the Energy Impact stays high after you stop dictating, the app may still be listening, so check its settings for a way to stop it.

Yap, which we make, runs its dictation on the device on Mac and Windows, so this question applies to it. Measure it with the steps above rather than relying on a figure from us. For the cost side, see whether on-device dictation is cheaper than cloud and why AI dictation apps charge a subscription. For setup, our guide to running Whisper locally explains how a local model is installed and what it needs.

This question belongs to our AI dictation cost guide, within the voice and AI overview. For the terms used here, see on-device speech recognition and Apple silicon in our glossary.