Feature request: Support custom wake words or match active profile persona name (e.g. "Hey Myra")

[Papyrus-Bundecca]

Problem

Currently, hardcodes the wake phrase to and strictly checks for this single phrase against the sherpa-onnx keyword spotter.

In multi-profile or customized Hermes deployments where the agent operates under a specific identity/persona (e.g. "Myra"), users expect to activate their assistant using that persona name rather than the generic project brand name.

Comments in already note that the routing schema was structured to support future profile-specific phrases:

"Stored now so a future profile-specific phrase UI can migrate without a schema rewrite. Product code intentionally writes Active in this release."

Proposed Solution

1. Configurable Wake Phrase in Settings:
   • Allow users to configure an arbitrary or selected wake phrase under Wake Word / Voice Settings (or infer "Hey " when a profile is pinned).
2. Dynamic sherpa-onnx Keyword Spotting:
   • Update / to generate or accept keyword spotting configurations / tokens for the chosen custom phrase instead of binding strictly to .
3. Fallback Grace Period:
   • Retain "Hey Hermes" as the fallback default while allowing the primary persona wake word to be recognized.

Context

Using Hermes Relay as the system default on Android allows full replacement of Google Assistant, and having custom wake phrase alignment completes the native assistant experience.

[hermes-relay-triage]

🤖 Hermes-Relay automated triage

Thanks for spelling out the profile-name use case! This is a confirmed current limitation rather than a hidden setting: the latest GitHub Release and current Android code intentionally support only the validated “Hey Hermes” phrase, and both background wake-listener paths use the same fixed sherpa-onnx keyword configuration. The stored profile-routing shape is future-facing; it does not currently provide custom phrase detection.

Supporting arbitrary or profile-derived names would require keyword generation plus routing and device acoustic validation, so this needs a product and implementation decision rather than a configuration workaround. A maintainer will follow up on whether the first supported scope should be a user-selected phrase or automatic “Hey ” behavior.