diff --git a/src/app/plugins/call/CallEmbed.ts b/src/app/plugins/call/CallEmbed.ts
index f029bfaec..7b2226765 100644
--- a/src/app/plugins/call/CallEmbed.ts
+++ b/src/app/plugins/call/CallEmbed.ts
@@ -237,10 +237,14 @@ export class CallEmbed {
         denoiseMode === 'browser' ||
         (denoiseMode === 'ml' && denoiseNativeNS)
       ).toString(),
-      // Turn the browser's auto gain control OFF for the ML tier only: its
-      // dynamic gain fights the in-source ML denoiser (pumping). Browser/off
-      // tiers keep the browser's normal capture pipeline (AGC on).
-      autoGainControl: (denoiseMode !== 'ml').toString(),
+      // Keep the browser's auto gain control ON in every tier. It used to be
+      // turned off for the ML tier ("pumping"), but AGC is what brings a quiet
+      // mic up to speaking level: without it ML-tier voices went out 10-20 dB
+      // quieter than the browser tier, while the models themselves keep speech
+      // within ~1.5 dB (measured: RNNoise, Speex, DTLN, DeepFilterNet3). Chrome's
+      // AGC2 only adapts during speech, and the model runs after it, so it still
+      // removes the noise AGC leaves in the pauses.
+      autoGainControl: 'true',
       audio: initialAudio.toString(),
       video: initialVideo.toString(),
       header: 'none',