Skip to content

fix(audeze): remove parameter-setting packet from Maxwell initialization - #577

Merged
Sapd merged 2 commits into
Sapd:masterfrom
NOVA2270:fix/audeze-maxwell-balance
Sep 10, 2026
Merged

Sapd merged 2 commits into
Sapd:masterfrom
NOVA2270:fix/audeze-maxwell-balance

Conversation

@NOVA2270

@NOVA2270 NOVA2270 commented Sep 9, 2026

Copy link
Copy Markdown
Contributor

Closes #561

This removes the parameter-setting packet from the initialization sequences of both Audeze Maxwell and Audeze Maxwell 2.

On my original Maxwell (0x3329:0x4b19), the packet caused a persistent left/right audio imbalance after a battery query. After removing it:

  • Battery reporting continues to work.
  • Repeated battery queries no longer change the audio balance.
  • The initialization request count was updated from 15 to 14.
  • The equivalent packet was also removed from the Maxwell 2 implementation, as requested.

Hardware testing was performed on the original Maxwell. I do not have a Maxwell 2 available for hardware testing.

@Sapd
Sapd merged commit 6eb2c99 into Sapd:master Sep 10, 2026
@Sapd Sapd mentioned this pull request Sep 10, 2026
2 tasks done
AdrianKuriata pushed a commit to AdrianKuriata/HeadsetControl-GUI that referenced this pull request Sep 11, 2026
Released headsetcontrol up to 4.1.0 sends a DSP parameter write on every info
read, and the app's 5 s refresh loop would send it constantly against a real
Maxwell. The target builds upstream master into .upstream/ and puts it first on
PATH; the script refuses to build if the packet ever reappears.

Temporary — delete it once Sapd/HeadsetControl#577 is in a release.

Claude-Session: https://claude.ai/code/session_01SkTQSQNwRB7RkCXW5coHHc
AdrianKuriata added a commit to AdrianKuriata/HeadsetControl-GUI that referenced this pull request Oct 16, 2026
* feat(detect): require headsetcontrol 4.1.0 (#67)

4.1.0 carries our upstream #549 and #550, so a parameter write costs 0.07 s
instead of 2.90 s on the Maxwell 2. The floor is now a latency one, not a
capability one — 4.0.0 works, just slowly — and ADR 0016 records that.

Fixtures re-recorded against 4.1.0: api_version is 1.5, a write lists devices
but reads no values from them, and the test device gained CAP_SIDETONE_STATUS
(rendered in #68). Reads come from a master build on purpose — released 4.1.0
still sends the packet that shifts a Maxwell's audio balance (upstream #577).

Install screens offer upstream's PPA/COPR/AUR first, packages second, source last.

Claude-Session: https://claude.ai/code/session_01SkTQSQNwRB7RkCXW5coHHc

* chore(dev): add make dev-upstream, an Audeze-safe dev run (#67)

Released headsetcontrol up to 4.1.0 sends a DSP parameter write on every info
read, and the app's 5 s refresh loop would send it constantly against a real
Maxwell. The target builds upstream master into .upstream/ and puts it first on
PATH; the script refuses to build if the packet ever reappears.

Temporary — delete it once Sapd/HeadsetControl#577 is in a release.

Claude-Session: https://claude.ai/code/session_01SkTQSQNwRB7RkCXW5coHHc

* fix(lint): keep the dev-upstream build out of ESLint and Prettier (#67)

CMake writes files named compiler_depend.ts under .upstream/, which the
type-aware ESLint run then tries to parse as TypeScript — 34 errors the moment
anyone runs make dev-upstream. .gitignore does not reach either tool.

Claude-Session: https://claude.ai/code/session_01SkTQSQNwRB7RkCXW5coHHc

---------

Co-authored-by: tinker <kuriatadev@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Bug: Audeze Maxwell audio balance shifts (right ear becomes louder) after battery check

2 participants