v2.0.0

Breaking change (Sendspin)

The Sendspin client was rebuilt on aiosendspin 9.x. The deprecated pre-encryption wire protocol and mDNS auto-discovery are gone, and the Music Assistant server address is now required in config.json:

"sendspin": { "connection": { "server_host": "<MA-SERVER-IP>" } }

Existing config.json files must add sendspin.connection.server_host before the client can connect. Run script/migrate_config.py after updating: it warns when the key is missing (it cannot guess your MA IP) and strips the other obsolete sendspin.connection keys. Devices without Sendspin enabled are unaffected. The rebuilt client adds Noise-encrypted pairing with a PIN spoken through the speaker (Piper, espeak-ng fallback), persistent player identity, and a sounddevice output stage with server-synchronized buffering. Requires Python >= 3.12 and libportaudio2.

Re-founded on the upstream core

This release lands the re-foundation: upstream modules (satellite.py, entity.py, models.py, player/, wake_word.py, peripheral API) are used as-is and the fork's differentiating features live as add-on modules. Future upstream releases merge cleanly again. From upstream this brings the full ESPHome device-page entity set (mic auto gain, noise suppression, mic volume, per-slot wake word and stop word sensitivities), dual music/TTS players with ducking and announcements, output-only mode, external wake word downloads, MWW/OWW model switching from the UI, and the WebSocket peripheral API (port 6055).

main is now the re-founded stack. Devices still tracking the old upstream-core branch keep working: their updater self-migrates to main on the first update after this release.

Kernel-independent ReSpeaker 2-Mic HAT audio (v1 and v2)

Audio runs on mainline kernel drivers (snd-soc-wm8960 for v1, snd-soc-tlv320aic3104 for v2) via device-tree overlays compiled at install time:

• No DKMS, no kernel headers, no per-kernel builds, any kernel >= 5.4; kernel upgrades can no longer break audio
• One smart installer auto-detects the HAT revision by I2C address (0x1a = v1, 0x18 = v2) and removes stale installs of the other revision
• Same ALSA card ID (seeed2micvoicec) as before, so existing LVA configs keep working; legacy DKMS installs upgrade in place
• v1 gets proper MCLK wiring on the codec node (fixes "No MCLK configured" PCM open failures), which also enables the WM8960 PLL for the 44.1 kHz family

Self-updating fleet

script/update_lva rewritten: in-place git fetch + checkout instead of move-aside + fresh clone. Untracked per-device files (preferences, Sendspin credentials, piper voices, per-model threshold files, external wake words, config.json) are never touched. The script re-executes the freshly fetched copy, so update logic always runs at the version being deployed. Adds --branch, --force, and --rollback. Now defaults to main.

Fixed

• Tray client stale state after MQTT (re)connects: the daemon now publishes a consolidated retained lva/<device_id>/state topic on every voice transition, and the tray derives its displayed state from it. State changes while muted are also published now.
• HA mute switch going stale on non-HA mute changes (tray, GPIO, XVF3800 hardware buttons): mute state is now published to all connected API clients.

Upgrade notes

1. Run script/migrate_config.py (or script/update_lva with your usual flags) after updating.
2. If you use Sendspin, add sendspin.connection.server_host to config.json.
3. config.json is no longer tracked by git; per-device files survive updates.

Full details in CHANGELOG.md.

Testing: 625 passed / 1 skipped with all install extras (Python 3.12); verified on Raspberry Pi Zero 2 W (2-Mic HAT v1 + v2), Orange Pi Zero 2 W (XVF3800), and desktop tray clients.