Windows desktop app for live production

XNN Translate

Live transcription and translated captions from audio on your PC, with local preview and output for OBS production workflows.

01  Overview

One applied system from XNN.

XNN Translate is one of XNN's applied systems — a complete application built on the company's underlying engineering.

It brings live transcription, translation, and captioning to the user's own machine, as a practical demonstration of that work outside the research environment.

02  Function

From live speech to readable captions.

XNN Translate listens to a selected local audio source, transcribes the speech, translates it into a chosen target language, and produces readable captions from the live session.

It keeps rolling context from the ongoing transcript and caption history so that translation reflects the conversation as it develops, rather than treating each phrase in isolation.

03  Workflow

Built for live production.

Captions are generated during the session and made available to streaming and recording workflows.

The application is designed to deliver caption output into live production software, including a local overlay source, and to retain a session transcript and translated caption history alongside the live output.

04  Local operation

Processed on your own hardware.

Speech recognition and translation run locally on the user's own machine. The live audio is not transmitted to a remote service to be transcribed or translated.

Processing the audio on the device keeps the source material on the machine where the session takes place.

05  Availability

Available for Windows.

The Download page reads the current release manifest and shows the available installer, version, file size, signing identity, and SHA-256 checksum.

The installer is code-signed by XNN LLC. A new release may still trigger a Windows SmartScreen reputation prompt; use the verification guidance shown with the release.

06  Requirements

Before you install.

XNN Translate is a 64-bit Windows desktop application. Speech recognition and translation run through a local processing backend selected for the available hardware.

Windows
Windows 10 version 22H2 or Windows 11, 64-bit.
Model files
First launch downloads about 3 GB of model files and stores them locally. Keep additional free disk space for application updates.
OBS output
OBS Studio 30 or newer for the guided workflow. Manual Browser Source and TXT output remain available as fallbacks.

GPU acceleration depends on compatible hardware and drivers. The application can use processor fallback when a usable GPU backend is unavailable.

07  Contact

Product help.

Installation, activation, OBS setup, and current service information are available from product support.

Open support →
← Back to XNN