One of the major issues with Anki going out of business is Vector voice recognition ending, rendering Vector deaf. Voice NLP currently uses a remote processing platform.
I've put the current english voice commands in a separate gist here: Vector voice commands
- The holy-grail version; no external dependencies.
- Device has significant processing power, can it be used for local voice recognition? (overheat issues?)
- Language-specific acoustic models are required on device, needs to be selectable and updateable by user, unless all languages fit.
- Rewrite the API calls to use a different API provider that similarly provides NLP.
- Least(?) amount of code to write.
- Minimize embedded code requirements.
- Requires on-network service.
- Authentication/pairing concerns.
- Discovery of service required (use mdns, like robot discovery?)