This fork is an attempt at implementing continued conversation and other features within Stream Assist.
Features added so far:
- Continued conversation: conversation_id is preserved between interactions for a given entry with a timeout of 300 seconds, though for now the timeout doesn't seem to work consistently
- Wake Word skip: The wake word phase can be automatically skipped on follow-up interactions
- Still a work in progress, very buggy
- STT End media: Now both the start and end of the STT phase have options for audio feedback. Also added a field for this in the config options
- STT Error media: When there is an error in the STT phase (like no-text-recognized), a distinct error sound can be played
- More initial config options: All available options are now in the initial config screen
- Added cancel phrases like: "nevermind", "never mind", "thank you", "cancel that", "cancel", "abort", "quit", "exit", "end", "forget it", "that's all", "that is all" so that you can cancel the continuous conversation feature
Future goals:
- Globally continued conversations: Add the option to pass conversation IDs across all integration entries
- This would require the integration to also provide updated area and device information
- Expose more of the integration to automations
- Example: A service call that allows you to select an integration entry and trigger the assistant pipeline at the intent phase with predefined data