diff --git a/src/voice/VoiceRecorder.ts b/src/voice/VoiceRecorder.ts
index 50497438ca..319a6c3a37 100644
--- a/src/voice/VoiceRecorder.ts
+++ b/src/voice/VoiceRecorder.ts
@@ -16,6 +16,7 @@ limitations under the License.
 
 import * as Recorder from 'opus-recorder';
 import encoderPath from 'opus-recorder/dist/encoderWorker.min.js';
+import mxVoiceWorkletPath from './mxVoiceWorklet';
 import {MatrixClient} from "matrix-js-sdk/src/client";
 import CallMediaHandler from "../CallMediaHandler";
 import {SimpleObservable} from "matrix-widget-api";
@@ -36,7 +37,7 @@ export class VoiceRecorder {
     private recorderSource: MediaStreamAudioSourceNode;
     private recorderStream: MediaStream;
     private recorderFFT: AnalyserNode;
-    private recorderProcessor: ScriptProcessorNode;
+    private recorderWorklet: AudioWorkletNode;
     private buffer = new Uint8Array(0);
     private mxc: string;
     private recording = false;
@@ -70,18 +71,20 @@ export class VoiceRecorder {
         // it makes the time domain less than helpful.
         this.recorderFFT.fftSize = 64;
 
-        // We use an audio processor to get accurate timing information.
-        // The size of the audio buffer largely decides how quickly we push timing/waveform data
-        // out of this class. Smaller buffers mean we update more frequently as we can't hold as
-        // many bytes. Larger buffers mean slower updates. For scale, 1024 gives us about 30Hz of
-        // updates and 2048 gives us about 20Hz. We use 1024 to get as close to perceived realtime
-        // as possible. Must be a power of 2.
-        this.recorderProcessor = this.recorderContext.createScriptProcessor(1024, CHANNELS, CHANNELS);
+        await this.recorderContext.audioWorklet.addModule(mxVoiceWorkletPath);
+        this.recorderWorklet = new AudioWorkletNode(this.recorderContext, "mx-voice-worklet");
 
         // Connect our inputs and outputs
         this.recorderSource.connect(this.recorderFFT);
-        this.recorderSource.connect(this.recorderProcessor);
-        this.recorderProcessor.connect(this.recorderContext.destination);
+        this.recorderSource.connect(this.recorderWorklet);
+        this.recorderWorklet.connect(this.recorderContext.destination);
+
+        // Dev note: we can't use `addEventListener` for some reason. It just doesn't work.
+        this.recorderWorklet.port.onmessage = (ev) => {
+            if (ev.data['ev'] === 'proc') {
+                this.tryUpdateLiveData(ev.data['timeMs']);
+            }
+        };
 
         this.recorder = new Recorder({
             encoderPath, // magic from webpack
@@ -128,7 +131,7 @@ export class VoiceRecorder {
         return this.mxc;
     }
 
-    private tryUpdateLiveData = (ev: AudioProcessingEvent) => {
+    private tryUpdateLiveData = (timeMillis: number) => {
         if (!this.recording) return;
 
         // The time domain is the input to the FFT, which means we use an array of the same
@@ -150,7 +153,7 @@ export class VoiceRecorder {
 
         this.observable.update({
             waveform: translatedData,
-            timeSeconds: ev.playbackTime,
+            timeSeconds: timeMillis / 1000,
         });
     };
 
@@ -166,7 +169,6 @@ export class VoiceRecorder {
         }
         this.observable = new SimpleObservable<IRecordingUpdate>();
         await this.makeRecorder();
-        this.recorderProcessor.addEventListener("audioprocess", this.tryUpdateLiveData);
         await this.recorder.start();
         this.recording = true;
     }
@@ -178,6 +180,7 @@ export class VoiceRecorder {
 
         // Disconnect the source early to start shutting down resources
         this.recorderSource.disconnect();
+        this.recorderWorklet.disconnect();
         await this.recorder.stop();
 
         // close the context after the recorder so the recorder doesn't try to
@@ -189,7 +192,6 @@ export class VoiceRecorder {
 
         // Finally do our post-processing and clean up
         this.recording = false;
-        this.recorderProcessor.removeEventListener("audioprocess", this.tryUpdateLiveData);
         await this.recorder.close();
 
         return this.buffer;
diff --git a/src/voice/mxVoiceWorklet.js b/src/voice/mxVoiceWorklet.js
new file mode 100644
index 0000000000..a74f5c17c9
--- /dev/null
+++ b/src/voice/mxVoiceWorklet.js
@@ -0,0 +1,35 @@
+/*
+Copyright 2021 The Matrix.org Foundation C.I.C.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+class MxVoiceWorklet extends AudioWorkletProcessor {
+    constructor() {
+        super();
+
+        this._timeStart = 0;
+    }
+
+    process(inputs, outputs, parameters) {
+        const now = (new Date()).getTime();
+        if (this._timeStart === 0) {
+            this._timeStart = now;
+        }
+
+        this.port.postMessage({ev: 'proc', timeMs: now - this._timeStart});
+        return true;
+    }
+}
+
+registerProcessor('mx-voice-worklet', MxVoiceWorklet);