On-device or cloud speech: an honest comparison

We get asked this a lot, usually by teams who already have an opinion. Both options are good. They are good at different things.

Where on-device wins

No network means no network delay and no outage when the train goes into a tunnel. Audio never leaves the phone, which matters for health and finance apps. And there is no per-character bill.

Where the cloud wins

The best voices are large. They need more memory than most phones can spare, and they get better every month without an app update. Custom and cloned voices also live more comfortably on servers.

The middle path

Many teams use both. A small on-device voice handles short system prompts and offline moments. The cloud handles long answers and anything with personality.

	On-device	Cloud
First audio	Instant	Under 200 ms
Voice quality	Good	Best
Works offline	Yes	No
Battery cost	Noticeable	Minimal
Custom voices	Limited	Yes

If you are unsure, start in the cloud. It is easier to move the simple parts onto the device later than to discover your voice sounds flat on half your users’ phones.