How to Speed Up Speaker Diarization
Speaker diarization is computationally intensive — but enabling GPU acceleration in Offline Transcriber can make it up to ~7× faster. This guide shows you how to turn on GPU support, which hardware is supported, and how to pick the right graphics adapter for your machine.
Why GPU Acceleration Matters
Diarization runs the same kind of neural-network workload as transcription itself, and graphics cards are built to run that workload many times faster than a CPU.
On our reference hardware (NVIDIA RTX 1080 Ti) a 12-minute audio file finishes diarization in about 1 minute on GPU vs ~7 minutes on CPU — roughly a 7× speedup. Newer or higher-tier GPUs can deliver even larger gains.
Heads up: The Graphics Adapter setting is global — it affects transcription, diarization, and other AI-powered features in Offline Transcriber. Switching it changes performance and behavior across the whole app, not just diarization.
Supported Hardware
How to Enable GPU Acceleration
-
Open Settings
Launch Offline Transcriber and open the Settings panel.
-
Go to Graphics Adapter
Navigate to Settings → Graphics Adapter. You'll see a list of every compatible GPU detected on your system, plus a "Not use GPU" option to fall back to CPU.
-
Choose Your Best Graphics Card
Select the most powerful GPU available. If your machine has both a discrete card (e.g. NVIDIA RTX, AMD Radeon RX) and an integrated GPU (Intel UHD, AMD Vega), always pick the discrete card — it will deliver substantially better performance.
-
Run Diarization as Usual
Once selected, GPU acceleration is active for all subsequent diarization (and transcription) runs. No per-job toggle is needed.
Choosing Between Multiple GPUs
If you have more than one supported GPU installed, Offline Transcriber lists each one separately under Graphics Adapter. A few rules of thumb when picking:
- Discrete over integrated: Always prefer your dedicated card over an integrated GPU.
- Newer generation beats older flagship: A newer mid-tier card often outperforms an older high-end card on ML workloads.
- More VRAM gives headroom: Diarization itself doesn't need much VRAM, but more headroom means you can keep other GPU apps running alongside.
When to Disable GPU
There are a few legitimate reasons to switch back to "Not use GPU":
- The GPU is needed elsewhere: Gaming, video editing, or other ML workloads on the same card.
- Driver issues: If transcription or diarization throws errors that disappear on CPU, fall back to CPU while you update drivers.
- Battery life on laptops: CPU may be more efficient on battery for short jobs.
Troubleshooting
My GPU isn't listed
On Windows, make sure your graphics driver is up to date and that you're on a modern version of Windows 10 or Windows 11 (DirectML requires modern Windows). On macOS, GPU acceleration requires an Apple Silicon Mac — Intel-based Macs are not supported and will fall back to CPU automatically.
Diarization isn't noticeably faster on GPU
First, confirm you actually selected a GPU under Settings → Graphics Adapter and didn't leave it on "Not use GPU". If multiple GPUs are listed, verify you've chosen the most powerful one (typically the discrete card, not integrated graphics).
I switched GPU and now my transcriptions behave differently
This is expected. The Graphics Adapter setting is global — it changes which device handles transcription, diarization, and other AI features. You don't need to switch it per task.
Privacy Note
GPU acceleration runs 100% locally, exactly like the CPU path. No audio data leaves your computer regardless of which device you select.