Windows application

VoxRelay

Turn audio playing on your Windows PC into floating English subtitles with local NVIDIA CUDA or CPU processing.

Get it from Microsoft Store Open Microsoft Store app
VoxRelay application logo
VoxRelay runs from the Windows notification area and displays only the subtitle text over other applications.

Local speech processing

Hear the computer. Read the conversation.

VoxRelay captures a Windows playback source through WASAPI loopback, recognizes speech locally, and keeps an always-on-top subtitle overlay available without opening a conventional app window.

  • Automatic spoken-language analysis every 30 seconds
  • Direct English transcription and speech-to-English translation
  • Source-language labels on translated subtitles
  • Resizable, movable floating text with no taskbar window
  • Notification-area source selection, visibility, and quit controls
  • Auto, NVIDIA CUDA, and CPU processing choices
  • Voice-activity, confidence, repetition, and common false-caption filtering
Read the Manual View FAQ Open Support

VoxRelay in use

Floating subtitles without a visible app window.

Open the Manual
VoxRelay displaying floating English subtitles over another Windows application
Clear English speech is transcribed locally and displayed as floating always-on-top text.
VoxRelay displaying an English subtitle translated from Spanish
Automatic mode translates supported non-English speech and marks the detected source language.
VoxRelay native notification-area menu with subtitle and source controls
The native notification-area menu controls visibility, playback-source selection, Auto/CUDA/CPU processing, source refresh, and exit.
VoxRelay floating subtitle area moved and resized on the Windows desktop
Move and resize the transparent subtitle area to fit the current Windows workspace.

Automatic mode

One overlay for transcription and translation.

30s

Language recheck

VoxRelay periodically analyzes confident speech and requires confirmation before switching the remembered language.

Local

Audio processing

Captured audio, transcription, language detection, and translation stay on the Windows computer.

Auto

GPU or CPU inference

Auto uses the larger CUDA model when a supported NVIDIA GPU is available and otherwise selects the smaller CPU model.

Requirements

Built for Windows playback audio.

  • 64-bit Windows 10 or Windows 11
  • A supported NVIDIA CUDA GPU with a current driver is optional for higher accuracy and lower latency
  • CPU processing uses the smaller Small/int8 model and has higher latency
  • An enabled Windows playback device that exposes loopback capture
  • Internet access for the initial speech-model download unless the model is already cached
  • Several gigabytes of free storage for the selected speech model and runtime

Privacy by architecture

Audio is not sent to Max Power IT.

VoxRelay does not require an account and contains no advertising, analytics, tracking, or telemetry SDK. The selected CUDA or CPU model is downloaded from Max Power IT over HTTPS, with Hugging Face retained as a backup source. Captured audio and subtitles are not included in model-download requests.

Read the Privacy Policy

VoxRelay is independently developed by Max Power IT LLC. NVIDIA and CUDA are trademarks and/or registered trademarks of NVIDIA Corporation in the United States and other countries. NVIDIA does not sponsor or endorse VoxRelay. Other names identify their respective products or services.