GUIDE
Choosing between on-device and on-premises translation
Check which languages you need, where network access is available, and how many people will use the system at once.
On-device processing uses language packs and models installed on each device. It can suit work without reliable network access, provided the hardware supports the required languages and functions. Check memory, storage, battery use, and response time on the device you intend to order.
With on-premises processing, devices connect to a server over your internal network. A server can support models too large for a handheld, but capacity still depends on its hardware and the number of simultaneous requests. Test network coverage at each place where the devices will be used.
List the languages needed for speech recognition, translation, and spoken output; support can differ between these functions. Then estimate simultaneous use and identify areas without network access. Your evaluation should also record where audio and text are processed and whether either is stored.
You can combine the two approaches: installed functions for offline use and a local server for other tasks when connected. Write down what remains available if the server or network cannot be reached. Do not assume a feature works offline because another feature on the device does.
Related: Vavus AI for phones · Concierge online