Ramp the microphone gate closed, so sentences stop ending in a click

[greluc]

The microphone half of the clicking, split out as promised in #412. Independent of it — I test-merged the two branches and they apply cleanly in either order.

The problem

With Mic Sensitivity on, the voice detector gates the microphone by writing gain.value directly:

onVoiceStop: () => {
    if (this.microphoneGain && … micSensitivityEnabled …) {
        this.microphoneGain.gain.value = 0;
    }

The zero is what hurts. At the moment the gate closes your signal is still mid-decay, so cutting it is a step the size of whatever you were saying — a click in every other player's ears at the end of every sentence you speak. You never hear it yourself; it happens in the stream you send, not the one you receive.

Scope is narrow, worth being clear about it: the gate only runs when micSensitivityEnabled is on and autoGainControl is off. With the shipped defaults the gain node isn't even created, so most users have never had this.

The change

Release ramped over 20 ms, attack left instant. Not symmetry for its own sake:

• Closing is where the step is large, because the signal is still going.
• Opening is where it is near zero by definition, so there is nothing to smooth — and a fade-in would soften the first consonant of every word, on top of the delay the detector already costs before it fires onVoiceStart. Word beginnings are the last thing to spend latency on.

Fast attack, ramped release, the way a noise gate is built.

Opening the gate now also cancels a release still in flight. That is a real bug in the current code as much as a consequence of the ramp: scheduled automation outlives a plain value write, so without the cancel, speaking again inside the ramp window would let the old ramp run on to zero and mute the word that had just re-opened the gate.

Why MicrophoneSoundBar is not touched

It mirrors the same gating for the settings level meter, so it looks like it needs the same fix. It doesn't: its ScriptProcessorNode reads inputBuffer to compute an RMS and never writes outputBuffer, so that graph outputs silence — the connect(ctx.destination) on it is only there to keep the processor running. There is no click to remove, and ramping it would make the meter lag the gate it exists to display.

Verification

npm run typecheck, npm run lint and npm run format:check pass.

Not verified by ear — I don't have a lobby, and this one really wants an ear rather than a reviewer. Two things I'd want someone to check: that the click at the end of a sentence is actually gone, and that 20 ms of release doesn't audibly smear the last syllable. If it does, the constant is the only thing that needs moving.

🤖 Generated with Claude Code

[github-actions]

Download the artifacts for this pull request:

• BetterCrewLink Unpacked 64bit.zip
• BetterCrewLink Installer Windows.zip
• BetterCrewLink Installer Linux.zip

---

This service is provided by nightly.link. These artifacts will expire in 90 days and will not be available for download after that time.

[OhMyGuus]

Thanks