Skip to content

Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS

fix([issue-12599]): emit voice:tts:cancel on barge-in speaking edge (#12719)

[._.]/549863414f58
diff --git a/server/sockets/voice.js b/server/sockets/voice.js
index 922c53c0ea..078ce4c2d3 100644
--- a/server/sockets/voice.js
+++ b/server/sockets/voice.js
@@ -11,7 +11,7 @@
// | voice:speak | voice:output:primary | voice:output:detached
// Call host (FaceTime Audio bridge):
// Inbound: voice:call:attach | voice:call:audio | voice:call:detach | voice:call:hangup
-// Outbound: voice:call:state | voice:call:tts
+// Outbound: voice:call:state | voice:call:tts | voice:tts:cancel
// Meeting capture (same page, same audio bridge, no LLM — STT only):
// Inbound: voice:capture:start | voice:call:audio | voice:capture:stop
// Outbound: voice:capture:state
@@ -563,13 +563,16 @@ export const registerVoiceHandlers = (socket) => {
const view = toInt16Samples(pcm);
if (asCallHost) {
+ const wasSpeaking = Boolean(call.endpointer.speaking);
const utterance = call.endpointer.push(view);
- // Barge-in: the caller talking over a reply cancels it, exactly as
- // pressing the widget's interrupt does.
- if (call.endpointer.speaking && call.busy) {
- call.ctrl?.abort();
- state.pendingDestructive = null;
- socket.emit('voice:interrupt', { reason: 'barge-in' });
+ // Barge-in: caller speech onset cancels playback on the host socket
+ // whether or not a reply turn is still actively generating.
+ if (!wasSpeaking && call.endpointer.speaking) {
+ socket.emit('voice:tts:cancel', { reason: 'barge-in' });
+ if (call.busy) {
+ call.ctrl?.abort();
+ state.pendingDestructive = null;
+ }
}
if (!utterance) return;
noteCallerSpeech();
diff --git a/server/sockets/voice.test.js b/server/sockets/voice.test.js
index 04096e7965..06037a87c2 100644
--- a/server/sockets/voice.test.js
+++ b/server/sockets/voice.test.js
@@ -700,11 +700,42 @@ describe('voice:capture socket handlers (meeting capture)', () => {
await flushMicrotasks();
expect(capturedSignal.aborted).toBe(true);
- expect(socket.emitted.some((e) => e.event === 'voice:interrupt' && e.payload.reason === 'barge-in')).toBe(true);
+ expect(socket.emitted.some((e) => e.event === 'voice:tts:cancel' && e.payload.reason === 'barge-in')).toBe(true);
+ expect(socket.emitted.some((e) => e.event === 'voice:interrupt')).toBe(false);
releaseFirstTurn({});
await flushMicrotasks();
});
+
+ it('caller-speech onset after runTurn has resolved emits voice:tts:cancel on speaking edge only', async () => {
+ const socket = makeFakeSocket();
+ registerVoiceHandlers(socket);
+ await socket.fire('voice:call:attach');
+
+ runTurn.mockResolvedValueOnce({ text: 'Hello there' });
+
+ // Complete a turn so call.busy returns to false
+ for (let i = 0; i < 14; i += 1) socket.fire('voice:call:audio', { pcm: loudFrame() });
+ for (let i = 0; i < 36; i += 1) socket.fire('voice:call:audio', { pcm: silentFrame() });
+ await flushMicrotasks();
+ expect(runTurn).toHaveBeenCalledTimes(1);
+
+ socket.emitted.length = 0;
+
+ // Caller begins speaking again while call.busy is false:
+ socket.fire('voice:call:audio', { pcm: loudFrame() });
+ await flushMicrotasks();
+
+ const cancels = socket.emitted.filter((e) => e.event === 'voice:tts:cancel' && e.payload?.reason === 'barge-in');
+ expect(cancels).toHaveLength(1);
+ expect(socket.emitted.some((e) => e.event === 'voice:interrupt')).toBe(false);
+
+ // Subsequent loud frame during the same utterance must not emit another cancel (onset edge only)
+ socket.fire('voice:call:audio', { pcm: loudFrame() });
+ await flushMicrotasks();
+ const cancelsAfterSecondFrame = socket.emitted.filter((e) => e.event === 'voice:tts:cancel' && e.payload?.reason === 'barge-in');
+ expect(cancelsAfterSecondFrame).toHaveLength(1);
+ });
});
describe('voice:call:audio frame alignment', () => {