Live Captions is one of Windows 11’s most useful accessibility features — until you open Task Manager and find it chewing through 20, 30, or even 40 percent of your CPU while your laptop’s fans scream. On paper, real-time transcription should be light work for a modern processor, and on Copilot+ PCs it should barely touch the CPU at all, thanks to the NPU.

High CPU usage from Live Captions usually has a specific, fixable cause: the NPU offload failing silently and dumping the model onto the CPU, a corrupted speech model looping on errors, translation running an extra model you don’t need, or third-party software — especially security suites — interfering with the audio pipeline. This guide walks you through diagnosing which one is hitting you and fixing it.

If your captions are also consuming excessive memory rather than CPU, our companion guide on Live Captions high memory usage covers that angle, since the two problems sometimes share a root cause. Battery drain is another common companion symptom — see Live Captions draining battery if your runtime has collapsed. And because third-party interference is such a frequent culprit, our piece on Bitdefender compatibility with ARM Windows is worth a read if you run that suite; on AMD machines, Ryzen AI Max NPU issues can also explain why the NPU offload never engages.

Why this happens

  • NPU offload silently failing: On Copilot+ PCs the caption model should run on the NPU. If the NPU driver glitches, Windows falls back to the CPU with no warning — and CPU inference is far more expensive.
  • Translation running a second model: Enabling caption translation loads an additional neural machine translation model alongside transcription, roughly doubling compute.
  • Corrupted or mismatched speech models: A damaged language pack can cause the recognizer to error and retry continuously, pinning a CPU core.
  • Antivirus pipeline interference: Security software scanning model files or hooking the audio capture path forces redundant processing on every audio chunk.
  • Microphone mode capturing everything: The “include microphone audio” option makes the engine transcribe ambient room noise all day, keeping the model at full load even during silence.
  • Outdated Windows speech components: Older cumulative updates shipped less-optimized caption models; newer builds are measurably more efficient.

How to fix it

Work through these in order — diagnose first, then act.

Fix 1: Confirm whether the NPU is doing the work

Open Task Manager (Ctrl + Shift + Esc), go to the Performance tab, and look for the NPU graph on a Copilot+ PC. Launch Live Captions with Win + Ctrl + L, play some audio with speech, and watch: NPU utilization should rise while CPU stays modest.

  1. Open Task Manager with Ctrl + Shift + Esc and go to Performance.
  2. Press Win + Ctrl + L to start Live Captions and play speech audio.
  3. Watch the NPU graph — it should show activity on Copilot+ PCs.
  4. Identify which process is consuming CPU in the Processes tab.
  5. Record baseline NPU vs CPU percentages for comparison.

Fix 2: Update NPU and chipset drivers

A failed NPU offload is most often a driver problem. Open Device Manager and check “Neural processors” for warning icons.

  1. Open Device Manager and inspect “Neural processors” for warnings.
  2. Download the latest NPU/chipset drivers from your laptop maker.
  3. Install the package and reboot the laptop.
  4. Re-run the Task Manager test with Live Captions active.
  5. Re-check Device Manager to confirm no warning icons remain.

Fix 3: Turn off translation if you don’t need it

Translation is the single biggest optional CPU cost in Live Captions. Every translated caption requires a second neural model pass, and on CPU fallback that cost is brutal.

  1. Open Live Captions settings via the gear icon.
  2. Check whether translation to another language is enabled.
  3. Switch translation off for everyday captioning.
  4. Watch Task Manager to confirm the CPU drop.
  5. Keep translation off by default and enable it only per video.

Fix 4: Disable microphone capture and trim the workload

The microphone option tells Live Captions to transcribe your room as well as system audio. In a noisy environment — keyboard clatter, a TV in the background, office chatter — the model never idles, processing sound continuously at full load.

  1. Open Live Captions settings and find the microphone option.
  2. Turn microphone capture off unless you need room transcription.
  3. Close apps generating audio you don’t need captioned.
  4. Reduce background noise if the microphone must stay on.
  5. Re-test with a noisy and a quiet room to compare load.

Fix 5: Add exclusions for your antivirus

Security suites are a classic hidden cause of caption CPU spikes. Real-time scanning can intercept the speech model’s file access or the audio pipeline itself, and on ARM Windows some suites run through emulation layers that multiply the overhead. If you need help finding exclusion settings, the Microsoft Windows support hub documents the process for Windows Security.

  1. Open your antivirus settings and find the exclusions list.
  2. Add exclusions for Windows speech and Live Captions components.
  3. Restart Live Captions with Win + Ctrl + L.
  4. Compare CPU usage against your baseline screenshot.
  5. Restart the laptop once so exclusions apply to all processes.

Fix 6: Repair the speech model and update Windows

If CPU usage stays high after the above, the speech model itself may be corrupted, causing endless error-retry loops. Go to Settings > Time & language > Language & region, remove the speech language pack, reboot, and add it back fresh — this downloads a clean model.

  1. Remove and re-add the speech language pack in Language settings.
  2. Reboot and let the fresh model download completely.
  3. Install all pending Windows cumulative updates.
  4. Re-test Live Captions CPU usage in Task Manager.

Closing

Live Captions should be a lightweight background feature, not a CPU hog. In most cases the fix is one of three things: getting the NPU offload working again with updated drivers, turning off unneeded translation, or stopping antivirus interference.

Frequently asked questions

Is it normal for Live Captions to use a lot of CPU?
Some CPU use is normal — real-time speech recognition is demanding. But sustained usage above 15-20% on a modern laptop, or enough to spin up fans constantly, signals a problem: usually a missing NPU offload, a stuck model, or conflicting software.
Should Live Captions run on the NPU instead of the CPU?
On Copilot+ PCs, yes — the NPU handles the caption model far more efficiently than the CPU. If Task Manager shows heavy CPU use instead, your NPU driver may be outdated or the offload may have silently failed.
Can antivirus software cause Live Captions CPU spikes?
Yes. Some security suites aggressively scan the speech-model files or intercept the audio pipeline, forcing repeated reprocessing. Adding the Live Captions components to exclusions often drops CPU usage immediately.
Will disabling Live Captions translation lower CPU usage?
Yes, noticeably. Translation runs a second AI model on top of transcription. If you only need captions in the original language, turning translation off can cut CPU usage substantially.