VoxShift is Windows voice translation software. Live translation with reference-voice speech generation requires an NVIDIA CUDA-capable GPU, Turing or newer with 4 GB VRAM or more. Speech recognition, translation, and voice generation run locally on your PC, and the requirements are the same for the Steam and Microsoft Store versions.
| Area | Requirement | Notes |
|---|---|---|
| OS | Windows 10 version 2004 or later / Windows 11 (64-bit) | Same requirements for the Steam and Microsoft Store versions |
| GPU / VRAM | NVIDIA CUDA GPU required for Live voice generation: Turing or newer (GeForce GTX 16 / RTX 20 series or newer), 4 GB VRAM or more | Recommended: RTX 30 series or newer (e.g. GeForce RTX 3060 12 GB). Radeon, Intel GPU, and no-GPU systems can run setup and diagnostics only; Live voice generation is unavailable |
| CPU | Intel Core i5-8400 / AMD Ryzen 5 2600 or better | Recommended: Intel Core i7-9700 / AMD Ryzen 7 3700X |
| RAM | 8 GB | 16 GB recommended |
| Audio | Microphone and audio output device | To send the translated voice to Discord, games, or OBS you need a virtual audio device (VB-CABLE). The Steam version can install it (optional); the Microsoft Store version guides you to the official VB-Audio page |
| Storage | 40 GB free space | 80 GB recommended; optional higher-quality models need more |
| Internet | For purchase/license and optional model downloads | The first-run models are included; speech recognition, translation, and voice generation run locally on your PC |
Reference-voice Live translation requires an NVIDIA CUDA GPU. On PCs with less GPU memory, VoxShift checks your PC and picks Low VRAM (about 3.8 GiB) or Ultra-low VRAM (about 2.4 GiB) automatically; Ultra-low VRAM translates on the CPU, so it can be slower. The Performance tab shows the estimated VRAM use for your setup.
If the microphone input, translated voice output, or virtual audio device is set wrong, Discord or OBS will not receive the translated speech.
The models needed for the first run are included. Optional higher-quality models are prepared in the Performance tab and need an extra download, more disk space, and, for voice generation, much more VRAM.
Games and streaming software can consume GPU/CPU resources, which affects VoxShift latency.
Requirements cannot be judged from a table alone. Test your own PC, audio devices, and target app before deciding to purchase.
Reference-voice Live translation requires an NVIDIA CUDA GPU, Turing or newer (GeForce GTX 16 / RTX 20 series or newer), with 4 GB VRAM or more; RTX 30 series or newer (e.g. RTX 3060 12 GB) is recommended. On Radeon, Intel GPU, or no-GPU systems, only setup and diagnostics work and Live voice generation is unavailable. Results also depend on settings and other apps running at the same time, so check with the free version on your own PC.
For GPUs with less memory there are Low VRAM (about 3.8 GiB) and Ultra-low VRAM (about 2.4 GiB) modes. VoxShift checks your PC and picks one automatically, and you can choose yourself under Performance → Manual settings. Ultra-low VRAM translates on the CPU, so it can be slower. An NVIDIA CUDA GPU is still required; test latency and quality before buying.
Not if you only listen yourself on headphones. To send the translated voice into Discord, game voice chat, or OBS as a microphone, you need a virtual audio device (VB-CABLE). The Steam version bundles the official VB-CABLE installer as an optional install; the Microsoft Store version does not bundle it and guides you to the official VB-Audio page instead.
