Right now you can only use hey_neon without creating a custom image. This should be configurable within the UI itself and users can upload the model(s) directly.
Probably requires that Iris/Hana have a persistent datastore, or we'd want to use a PVC in a location that the skill uploads the information to. If we use XDG it would support non-containerized use cases as well. We may be able to store it in the browser cache but that's not a great place long-term.
Initial Implementation Requirements
Users can upload an OWW tflite or onnx file and specify the name in text, and the UI will switch to use that wakeword.
Wakeword settings must persist past container reboots.
Objective
Right now you can only use hey_neon without creating a custom image. This should be configurable within the UI itself and users can upload the model(s) directly.
Probably requires that Iris/Hana have a persistent datastore, or we'd want to use a PVC in a location that the skill uploads the information to. If we use XDG it would support non-containerized use cases as well. We may be able to store it in the browser cache but that's not a great place long-term.
Initial Implementation Requirements
Users can upload an OWW tflite or onnx file and specify the name in text, and the UI will switch to use that wakeword. Wakeword settings must persist past container reboots.
Other Considerations
No response