Skip to content

Wake word "Hi, StackChan" not recognized + Request for custom wake word support #94

Open

Description

@narutaro

Problem

The default wake word "Hi, StackChan" does not respond reliably. Tapping the screen works fine, confirming the microphone is functional.

The pronunciation of "StackChan" varies significantly across regions and languages. Japanese speakers naturally say "スタックチャン" which differs from how
the Wakenet model may have been trained.

Observation from source code

Kconfig.projbuild already defines USE_CUSTOM_WAKE_WORD with Multinet support, including configurable threshold. However:

  1. This is a compile-time option, not user-configurable at runtime
  2. The custom wake word input uses pinyin (Chinese phonetics), which doesn't help non-Chinese users
  3. The official documentation states "custom wake word support is not yet available"

Requested improvements

  1. Short term: Improve recognition of "Hi, StackChan" for Japanese speakers, or allow sensitivity adjustment (threshold is already in Kconfig as
    CUSTOM_WAKE_WORD_THRESHOLD, default 20%)
  2. Long term: Expose custom wake word configuration to users via the StackChan World app, supporting non-Chinese phonetics

Environment

  • Device: StackChan (Kickstarter edition)
  • Firmware: v1.4.2 (latest)
  • Language setting: Japanese
  • Microphone test: Passed
  • Screen tap wake: Works correctly
  • Tested in quiet room, various distances and volumes

Reference

  • firmware/main/Kconfig.projbuild line 687-745: Wake word configuration
  • Official docs state: "custom wake word support is not yet available"

Activity

  1. alfem commented on Jun 15, 2026

    @alfem

    As non english native, my "Hi Stackchan" pronuntiation is hard to understand by my companion.

    I'd suggest to give users three pre-compiled alternatives, like Google does (OK google, Hey Google). A community poll would be nice ;-)

  2. Forairaaaaa commented on Oct 28, 2026

    @Forairaaaaa
    Contributor

    Thanks for the detailed report. The default “Hi, StackChan” uses a pretrained WakeNet model. CUSTOM_WAKE_WORD_THRESHOLD applies to the separate MultiNet implementation, so changing it will not adjust the default wake-word detector.

    Recognition of Japanese pronunciation needs further evaluation, and user-configurable wake words are not currently supported. We’ll keep this issue open to track these requests. In the meantime, tapping the screen remains available as an alternative.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions