3 ms·
The performance of such a system would be bad compared to an internet connected one (frequent speech model updates and more compute) and the reviews would pillo
by yellow_postit 7y ago
The performance of such a system would be bad compared to an internet connected one (frequent speech model updates and more compute) and the reviews would pillory the device for a nebulous to most privacy (unseen) vs. performance (seen) trade off. Try using a slightly older (early 00s) car with voice commands to get a sense.
- zitterbewegung 7y agoYou can separate the recognition system of the speech model by allowing software updates but not send the data or the captured audio to the cloud.
- cortesoft 7y agoCouldn't a software update then change that behavior to send the data to the cloud?
- zitterbewegung 7y agoYes, this could happen but if the project was sending data to the cloud being opt in instead of opt out (like for example telemetry for Windows / MacOS when you first install) it would ideally be that way.