@burdenoff/fe-libs 2026.1028.2 → 2026.1028.4

package/dist/shared/assistant/ui/useVoiceSession.js

@@ -4,33 +4,37 @@ import { readHandsFreeOverride as r, resolveHandsFreeEnabled as ee } from "../vo
 import { silencePlaybackReference as te } from "../voice/handsfree/playbackReference.js";
 import { primeHostedPlayback as ne } from "../speech/hostedTtsSpeechEngine.js";
 import { speechFeed as re } from "../speech/speechFeed.js";
-import { INITIAL_SESSION as ie, INTERRUPT_MAX_MS as ae, NOTHING_HEARD as i, advanceHeardClock as oe, isCapturing as a, isSubmittable as se, planListeningHold as ce, planReplyConsumption as le, planVoiceAnnouncement as ue, shouldRecognise as de, spokenSoFar as o, voiceSessionReducer as fe } from "../voice/voiceSession.js";
-import { useHandsFreeInterrupt as pe } from "../voice/handsfree/useHandsFreeInterrupt.js";
-import { isTurnSuperseded as me } from "../voice/supersededTurns.js";
-import { captionChunks as he } from "../voice/speechPlayback.js";
-import { turnClockMs as s } from "../turnTimeline.js";
-import { useAssistantSpeech as ge } from "./AssistantSpeechProvider.js";
-import { useLiveAvatarOn as _e } from "./assistantLiveAvatarSeam.js";
-import { useCallback as c, useEffect as l, useLayoutEffect as ve, useMemo as ye, useReducer as be, useRef as u, useState as d } from "react";
+import { INITIAL_SESSION as ie, INTERRUPT_MAX_MS as ae, NOTHING_HEARD as oe, advanceHeardClock as se, isCapturing as ce, isSubmittable as le, planListeningHold as ue, planReplyConsumption as de, planVoiceAnnouncement as fe, shouldRecognise as pe, spokenSoFar as i, voiceSessionReducer as me } from "../voice/voiceSession.js";
+import { useHandsFreeInterrupt as he } from "../voice/handsfree/useHandsFreeInterrupt.js";
+import { isTurnSuperseded as ge } from "../voice/supersededTurns.js";
+import { captionChunks as _e } from "../voice/speechPlayback.js";
+import { turnClockMs as a } from "../turnTimeline.js";
+import { useAssistantSpeech as ve } from "./AssistantSpeechProvider.js";
+import { useLiveAvatarOn as ye } from "./assistantLiveAvatarSeam.js";
+import { useCallback as o, useEffect as s, useLayoutEffect as be, useMemo as xe, useReducer as Se, useRef as c, useState as l } from "react";
 //#region src/shared/assistant/ui/useVoiceSession.ts
-var xe = 4e3, Se = 2500, Ce = 5e3, we = 12e3;
-function Te({ onSubmit: Te, latestReply: f, language: p, graceMs: Ee, onSupersede: De, handsFreeInterrupt: Oe, handsFreeStreamNoticeAccepted: ke, onAcceptHandsFreeStreamNotice: Ae, onBargeIn: je }) {
-	let Me = n(Ee), m = u(null), [h, g] = be(c((e, t) => fe(e, t, m.current), []), ie), _ = ge(), { setSessionSpeaking: v, setSessionMicOpen: y, speakingMessageId: b, stop: x } = _, S = u(_.toggle), C = _.toggle;
-	ve(() => {
-		S.current = C;
-	}, [C]);
-	let w = u(null), T = u({}), E = u(i), D = u(null), O = u(!1), k = u(null), [Ne, A] = d(null), [j, M] = d(!1), [Pe, N] = d(!1), [P, F] = d(!1), I = u(h);
-	I.current = h;
-	let L = u([]), R = u(null), z = u(!1), B = c(() => {
+var Ce = 4e3, we = 2500, Te = 5e3, Ee = 12e3;
+function De({ onSubmit: De, latestReply: u, language: d, graceMs: Oe, onSupersede: ke, handsFreeInterrupt: Ae, handsFreeStreamNoticeAccepted: je, onAcceptHandsFreeStreamNotice: Me, onBargeIn: Ne }) {
+	let f = n(Oe), p = c(null), [m, h] = Se(o((e, t) => me(e, t, p.current), []), ie), g = ve(), { setSessionSpeaking: _, setSessionMicOpen: v, speakingMessageId: y, stop: b } = g, x = c(g.toggle), Pe = g.toggle;
+	be(() => {
+		x.current = Pe;
+	}, [Pe]);
+	let S = c(null), C = c({}), w = c(oe), T = c(void 0), E = c(void 0);
+	s(() => {
+		(m.phase === "idle" || m.phase === "ended") && (T.current = void 0, E.current = void 0);
+	}, [m.phase]);
+	let D = c(null), O = c(!1), k = c(null), [Fe, A] = l(null), [j, M] = l(!1), [Ie, N] = l(!1), [P, F] = l(!1), I = c(m);
+	I.current = m;
+	let L = c([]), R = c(null), z = c(!1), B = o(() => {
 		L.current = [], R.current = null, z.current = !1;
-	}, []), V = c(() => {
+	}, []), V = o(() => {
 		for (;;) {
 			let e = L.current.shift();
 			if (!e) return !1;
 			if (!(e.stillWanted && !e.stillWanted())) {
-				R.current = e, m.current = m.current ? `${m.current} ${e.text}` : e.text;
+				R.current = e, p.current = p.current ? `${p.current} ${e.text}` : e.text;
 				try {
-					S.current(e.id, e.text, e.lang);
+					x.current(e.id, e.text, e.lang);
 				} catch {
 					R.current = null;
 					continue;
@@ -38,50 +42,50 @@ function Te({ onSubmit: Te, latestReply: f, language: p, graceMs: Ee, onSupersed
 				return !0;
 			}
 		}
-	}, []), H = u(!1), Fe = c((e) => {
-		let t = ce({
+	}, []), H = c(!1), Le = o((e) => {
+		let t = ue({
 			requested: e,
 			held: H.current,
 			muted: I.current.muted
 		});
-		H.current = t.held, t.mute !== null && g({
+		H.current = t.held, t.mute !== null && h({
 			type: "mute",
 			muted: t.mute
 		});
-	}, []), U = c((e) => {
-		let t = ue(I.current);
-		t !== "drop" && (R.current?.id === e.id || L.current.some((t) => t.id === e.id) || (L.current.push(e), t === "start" && !z.current && (z.current = !0, g({ type: "announce" }))));
-	}, []), Ie = c((e) => {
+	}, []), U = o((e) => {
+		let t = fe(I.current);
+		t !== "drop" && (R.current?.id === e.id || L.current.some((t) => t.id === e.id) || (L.current.push(e), t === "start" && !z.current && (z.current = !0, h({ type: "announce" }))));
+	}, []), Re = o((e) => {
 		if (I.current.phase !== "thinking") {
 			U(e);
 			return;
 		}
-		L.current.push(e), z.current = !0, g({ type: "answered-locally" });
-	}, [U]), Le = u(f);
-	Le.current = f, l(() => {
-		if (h.phase === "speaking") {
+		L.current.push(e), z.current = !0, h({ type: "answered-locally" });
+	}, [U]), ze = c(u);
+	ze.current = u, s(() => {
+		if (m.phase === "speaking") {
 			let e = k.current;
-			m.current = e && D.current === e.id && f?.id !== e.id ? e.text : f?.text ?? null;
+			p.current = e && D.current === e.id && u?.id !== e.id ? e.text : u?.text ?? null;
 			return;
 		}
-		if (m.current === null) return;
+		if (p.current === null) return;
 		let e = setTimeout(() => {
-			m.current = null;
-		}, Se);
+			p.current = null;
+		}, we);
 		return () => clearTimeout(e);
-	}, [h.phase, f]);
-	let W = u(p);
-	W.current = p;
-	let G = c((e, t) => {
-		g({
+	}, [m.phase, u]);
+	let W = c(d);
+	W.current = d;
+	let G = o((e, t) => {
+		h({
 			type: "speech",
 			transcript: e,
 			final: t,
 			at: Date.now(),
 			language: W.current
 		});
-	}, []), Re = c((e, t, n) => {
-		g({
+	}, []), Be = o((e, t, n) => {
+		h({
 			type: "speech",
 			transcript: e,
 			final: t,
@@ -89,176 +93,185 @@ function Te({ onSubmit: Te, latestReply: f, language: p, graceMs: Ee, onSupersed
 			language: n ?? W.current,
 			gated: !0
 		});
-	}, []), K = u(0), ze = c(() => {
-		B(), x(), K.current = Date.now(), M(!0), g({ type: "interrupt" });
-	}, [x, B]), q = e({
-		onFinalize: c((e) => G(e, !0), [G]),
-		language: p
-	}), Be = u(je);
-	Be.current = je;
-	let Ve = c(() => {
-		B(), te(), x(), K.current = Date.now(), M(!0), g({ type: "interrupt" }), Be.current?.();
-	}, [x, B]), He = _e(), J = pe({
-		enabled: ee(Oe, r(), He),
-		phase: h.phase,
-		muted: h.muted,
-		language: p,
-		streamNoticeAccepted: ke ?? null,
-		onAcceptStreamNotice: Ae,
-		onBargeIn: Ve,
-		onTranscript: Re
-	}), Ue = J.mode === "notice", Y = q.isListening || J.suppressWebSpeech && J.mode === "active";
-	l(() => {
+	}, []), K = c(0), Ve = o(() => {
+		B(), b(), K.current = Date.now(), M(!0), h({ type: "interrupt" });
+	}, [b, B]), q = e({
+		onFinalize: o((e) => G(e, !0), [G]),
+		language: d,
+		onSpeechEnd: (e) => {
+			I.current.phase === "listening" && (E.current = e);
+		}
+	}), He = c(Ne);
+	He.current = Ne;
+	let Ue = o(() => {
+		B(), te(), b(), K.current = Date.now(), M(!0), h({ type: "interrupt" }), He.current?.();
+	}, [b, B]), We = ye(), J = he({
+		enabled: ee(Ae, r(), We),
+		phase: m.phase,
+		muted: m.muted,
+		language: d,
+		streamNoticeAccepted: je ?? null,
+		onAcceptStreamNotice: Me,
+		onBargeIn: Ue,
+		onTranscript: Be,
+		onSpeechFrame: (e) => {
+			I.current.phase === "listening" && !I.current.muted && (T.current = e);
+		}
+	}), Ge = J.mode === "notice", Y = q.isListening || J.suppressWebSpeech && J.mode === "active";
+	s(() => {
 		q.interim && G(q.interim, !1);
-	}, [q.interim, G]), l(() => {
-		w.current && clearTimeout(w.current), N(!1), E.current = oe(E.current, h, s());
+	}, [q.interim, G]), s(() => {
+		S.current && clearTimeout(S.current), N(!1), w.current = se(w.current, m, a());
 		let e = t({
-			phase: h.phase,
-			muted: h.muted,
-			spoken: o(h),
+			phase: m.phase,
+			muted: m.muted,
+			spoken: i(m),
 			holding: P,
-			graceMs: Me
+			graceMs: f
 		});
 		if (!e.arm) return;
-		let n = setTimeout(() => N(!0), Math.min(800, e.delayMs)), r = s();
-		return w.current = setTimeout(() => {
-			T.current = {
-				speechEnded: E.current.at ?? r,
+		let n = setTimeout(() => N(!0), Math.min(800, e.delayMs)), r = a();
+		return S.current = setTimeout(() => {
+			C.current = {
+				speechEnded: w.current.at ?? r,
+				lastSpeechFrame: T.current,
+				recognizerSpeechEnded: E.current,
+				selectedGraceMs: f,
 				graceStarted: r,
-				graceEnded: s()
-			}, g({
+				graceEnded: a()
+			}, h({
 				type: "silence",
 				at: Date.now()
 			});
 		}, e.delayMs), () => {
-			clearTimeout(n), w.current && clearTimeout(w.current);
+			clearTimeout(n), S.current && clearTimeout(S.current);
 		};
 	}, [
-		h,
+		m,
 		P,
-		Me
-	]), l(() => {
-		(h.phase !== "listening" || h.muted) && F(!1);
-	}, [h.phase, h.muted]), l(() => {
+		f
+	]), s(() => {
+		(m.phase !== "listening" || m.muted) && F(!1);
+	}, [m.phase, m.muted]), s(() => {
 		if (!j) return;
 		let e = Date.now() - K.current, t = Y ? Math.max(0, 500 - e) : Math.max(0, ae - e), n = setTimeout(() => M(!1), t);
 		return () => clearTimeout(n);
 	}, [j, Y]);
-	let We = u("");
-	h.phase === "listening" && (We.current = o(h)), l(() => {
-		if (h.phase !== "thinking") return;
-		y(!1), B(), A((e) => typeof e == "object" && e ? null : e);
-		let e = We.current.trim();
-		e && Te(e, Q.current ?? `vs-orphan-${Date.now().toString(36)}`, T.current, { correction: I.current.correcting }), T.current = {}, E.current = i;
-	}, [h.phase]), l(() => {
-		let e = le(f?.id, h.phase, D.current, me(f?.id));
-		e.claim && (D.current = f?.id ?? null, e.advance && f && (f.refused ? (B(), A({
+	let Ke = c("");
+	m.phase === "listening" && (Ke.current = i(m)), s(() => {
+		if (m.phase !== "thinking") return;
+		v(!1), B(), A((e) => typeof e == "object" && e ? null : e);
+		let e = Ke.current.trim();
+		e && De(e, Q.current ?? `vs-orphan-${Date.now().toString(36)}`, C.current, { correction: I.current.correcting }), C.current = {}, w.current = oe, T.current = void 0, E.current = void 0;
+	}, [m.phase]), s(() => {
+		let e = de(u?.id, m.phase, D.current, ge(u?.id));
+		e.claim && (D.current = u?.id ?? null, e.advance && u && (u.refused ? (B(), A({
 			kind: "turn-refused",
-			text: f.text
-		}), g({ type: "turn-refused" })) : g({
+			text: u.text
+		}), h({ type: "turn-refused" })) : h({
 			type: "reply",
-			text: f.text
+			text: u.text
 		})));
 	}, [
-		f,
-		h.phase,
+		u,
+		m.phase,
 		B
-	]), l(() => {
-		if (b === null || h.phase !== "thinking" || D.current === b) return;
-		if (me(b)) {
-			x();
+	]), s(() => {
+		if (y === null || m.phase !== "thinking" || D.current === y) return;
+		if (ge(y)) {
+			b();
 			return;
 		}
-		let e = re.peek(b);
-		!e || e.channel !== "voice" || e.voiceSessionId === void 0 || e.voiceSessionId !== Q.current || (D.current = b, k.current = {
-			id: b,
+		let e = re.peek(y);
+		!e || e.channel !== "voice" || e.voiceSessionId === void 0 || e.voiceSessionId !== Q.current || (D.current = y, k.current = {
+			id: y,
 			text: e.text
-		}, g({
+		}, h({
 			type: "reply-audio-started",
-			replyId: b
+			replyId: y
 		}));
 	}, [
-		b,
-		h.phase,
-		x
-	]), l(() => {
-		let e = b !== null;
-		O.current && !e && (R.current = null, (I.current.phase !== "speaking" || !V()) && g({ type: "speech-ended" })), O.current = e;
-	}, [b, V]), l(() => {
-		h.phase !== "listening" && A((e) => e === "nothing-heard" ? null : e);
-	}, [h.phase]), l(() => {
-		if (h.phase !== "speaking" || b !== null) return;
+		y,
+		m.phase,
+		b
+	]), s(() => {
+		let e = y !== null;
+		O.current && !e && (R.current = null, (I.current.phase !== "speaking" || !V()) && h({ type: "speech-ended" })), O.current = e;
+	}, [y, V]), s(() => {
+		m.phase !== "listening" && A((e) => e === "nothing-heard" ? null : e);
+	}, [m.phase]), s(() => {
+		if (m.phase !== "speaking" || y !== null) return;
 		let e = setTimeout(() => {
-			O.current || (B(), g({ type: "speech-ended" }));
-		}, xe);
+			O.current || (B(), h({ type: "speech-ended" }));
+		}, Ce);
 		return () => clearTimeout(e);
 	}, [
-		h.phase,
-		b,
+		m.phase,
+		y,
 		B
 	]);
-	let X = _.outputNotCleared;
-	l(() => {
+	let X = g.outputNotCleared;
+	s(() => {
 		if (X === null || D.current !== X.messageId) return;
 		let { phase: e } = I.current;
 		e === "idle" || e === "ended" || e === "thinking" || A("reply-not-heard");
-	}, [X]), l(() => {
-		h.phase === "listening" || h.phase === "thinking" || A((e) => e === "reply-not-heard" ? null : e);
-	}, [h.phase]);
-	let Z = u(De);
-	Z.current = De;
-	let Ge = u(0);
-	l(() => {
-		let e = h.superseded;
-		e !== Ge.current && (Ge.current = e, e !== 0 && (B(), x(), Z.current?.()));
+	}, [X]), s(() => {
+		m.phase === "listening" || m.phase === "thinking" || A((e) => e === "reply-not-heard" ? null : e);
+	}, [m.phase]);
+	let Z = c(ke);
+	Z.current = ke;
+	let qe = c(0);
+	s(() => {
+		let e = m.superseded;
+		e !== qe.current && (qe.current = e, e !== 0 && (B(), b(), Z.current?.()));
 	}, [
-		h.superseded,
+		m.superseded,
 		B,
-		x
-	]), l(() => {
-		y(a(h)), de(h) && !J.suppressWebSpeech ? q.isActive || q.start("auto") : q.isActive && q.stop();
+		b
+	]), s(() => {
+		v(ce(m)), pe(m) && !J.suppressWebSpeech ? q.isActive || q.start("auto") : q.isActive && q.stop();
 	}, [
-		h,
+		m,
 		q,
-		y,
+		v,
 		J.suppressWebSpeech
-	]), l(() => {
-		h.phase !== "speaking" || !z.current || (z.current = !1, V() || g({ type: "speech-ended" }));
-	}, [h.phase, V]), l(() => {
-		q.error === "not-allowed" || q.error === "audio-capture" ? g({
+	]), s(() => {
+		m.phase !== "speaking" || !z.current || (z.current = !1, V() || h({ type: "speech-ended" }));
+	}, [m.phase, V]), s(() => {
+		q.error === "not-allowed" || q.error === "audio-capture" ? h({
 			type: "fail",
 			error: "mic-denied"
-		}) : q.error === "language-not-supported" ? g({
+		}) : q.error === "language-not-supported" ? h({
 			type: "fail",
 			error: "no-speech-support"
 		}) : q.error === "network" ? A("network") : q.error === "unknown" && A("recognition-failed");
-	}, [q.error]), l(() => {
+	}, [q.error]), s(() => {
 		q.error !== null || !Y || A((e) => e === "network" || e === "recognition-failed" ? null : e);
-	}, [q.error, Y]), l(() => {
-		if (Y && A((e) => e === "not-capturing" ? null : e), !a(h) || Y || Ue) return;
-		let e = setTimeout(() => A("not-capturing"), Ce);
+	}, [q.error, Y]), s(() => {
+		if (Y && A((e) => e === "not-capturing" ? null : e), !ce(m) || Y || Ge) return;
+		let e = setTimeout(() => A("not-capturing"), Te);
 		return () => clearTimeout(e);
 	}, [
-		h,
+		m,
 		Y,
-		Ue
-	]), l(() => {
-		if (h.phase !== "listening" || h.muted || !Y || o(h) !== "") return;
-		let e = setTimeout(() => A("nothing-heard"), we);
+		Ge
+	]), s(() => {
+		if (m.phase !== "listening" || m.muted || !Y || i(m) !== "") return;
+		let e = setTimeout(() => A("nothing-heard"), Ee);
 		return () => clearTimeout(e);
-	}, [h, Y]), l(() => {
-		o(h) !== "" && A(null);
-	}, [h]);
-	let Q = u(null), Ke = c(() => {
+	}, [m, Y]), s(() => {
+		i(m) !== "" && A(null);
+	}, [m]);
+	let Q = c(null), Je = o(() => {
 		if (M(!1), N(!1), F(!1), B(), H.current = !1, !q.isSupported) {
-			g({
+			h({
 				type: "fail",
 				error: "no-speech-support"
 			});
 			return;
 		}
-		if (_.isBlocked) {
-			g({
+		if (g.isBlocked) {
+			h({
 				type: "fail",
 				error: "no-voice-support"
 			});
@@ -271,54 +284,57 @@ function Te({ onSubmit: Te, latestReply: f, language: p, graceMs: Ee, onSupersed
 				t.volume = 0, e.speak(t);
 			}
 		} catch {}
-		ne(), v(!0), A(null), D.current = Le.current?.id ?? null, k.current = null, Q.current = `vs-${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 8)}`, g({ type: "start" });
+		ne(), _(!0), A(null), D.current = ze.current?.id ?? null, k.current = null, Q.current = `vs-${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 8)}`, h({ type: "start" });
 	}, [
 		q.isSupported,
-		_.isBlocked,
-		v,
+		g.isBlocked,
+		_,
 		B
-	]), qe = c(() => {
-		M(!1), N(!1), F(!1), B(), H.current = !1, Q.current = null, g({ type: "end" }), x(), v(!1);
+	]), Ye = o(() => {
+		M(!1), N(!1), F(!1), B(), H.current = !1, Q.current = null, h({ type: "end" }), b(), _(!1);
 	}, [
-		x,
-		v,
+		b,
+		_,
 		B
-	]), Je = c((e) => Q.current === e, []), Ye = c(() => {
+	]), Xe = o((e) => Q.current === e, []), Ze = o(() => {
 		let e = I.current;
 		e.phase !== "listening" || e.muted || F(!0);
-	}, []), Xe = c(() => {
+	}, []), Qe = o(() => {
 		F(!1);
 		let e = I.current;
-		if (e.phase !== "listening" || e.muted || !se(o(e))) return;
-		let t = s();
-		T.current = {
-			speechEnded: E.current.at ?? t,
+		if (e.phase !== "listening" || e.muted || !le(i(e))) return;
+		let t = a();
+		C.current = {
+			speechEnded: w.current.at ?? t,
+			lastSpeechFrame: T.current,
+			recognizerSpeechEnded: E.current,
+			selectedGraceMs: f,
 			graceStarted: t,
 			graceEnded: t
-		}, g({
+		}, h({
 			type: "silence",
 			at: Date.now()
 		});
-	}, []), Ze = c(() => {
-		g({
+	}, [f]), $e = o(() => {
+		h({
 			type: "mute",
 			muted: !I.current.muted
 		});
 	}, []);
-	l(() => () => {
-		v(!1), y(!1);
-	}, [v, y]);
-	let $ = he(_.spokenChunks, h.phase, b);
-	return ye(() => ({
-		session: h,
-		notice: Ne ?? (J.fallbackReason !== null && J.fallbackReason !== "notice-declined" ? "hands-free-fallback" : J.nothingHeard ? "hands-free-nothing-heard" : null),
+	s(() => () => {
+		_(!1), v(!1);
+	}, [_, v]);
+	let $ = _e(g.spokenChunks, m.phase, y);
+	return xe(() => ({
+		session: m,
+		notice: Fe ?? (J.fallbackReason !== null && J.fallbackReason !== "notice-declined" ? "hands-free-fallback" : J.nothingHeard ? "hands-free-nothing-heard" : null),
 		spokenChunks: $,
 		capturing: Y,
-		active: h.phase !== "idle" && (h.phase !== "ended" || h.error !== null),
+		active: m.phase !== "idle" && (m.phase !== "ended" || m.error !== null),
 		interrupting: j,
-		awaitingEndpoint: Pe,
+		awaitingEndpoint: Ie,
 		holding: P,
-		correcting: h.correcting,
+		correcting: m.correcting,
 		handsFree: {
 			mode: J.mode,
 			nothingHeard: J.nothingHeard,
@@ -326,22 +342,22 @@ function Te({ onSubmit: Te, latestReply: f, language: p, graceMs: Ee, onSupersed
 			fallbackReason: J.fallbackReason,
 			privacyNotice: J.privacyNotice
 		},
-		available: q.isSupported && !_.isBlocked,
-		start: Ke,
-		end: qe,
-		isCurrentSession: Je,
-		toggleMute: Ze,
-		interrupt: ze,
-		letMeFinish: Ye,
-		imDone: Xe,
+		available: q.isSupported && !g.isBlocked,
+		start: Je,
+		end: Ye,
+		isCurrentSession: Xe,
+		toggleMute: $e,
+		interrupt: Ve,
+		letMeFinish: Ze,
+		imDone: Qe,
 		announce: U,
-		holdListening: Fe,
-		answerLocally: Ie
+		holdListening: Le,
+		answerLocally: Re
 	}), [
-		h,
-		Ne,
+		m,
+		Fe,
 		j,
-		Pe,
+		Ie,
 		P,
 		Y,
 		J.mode,
@@ -350,19 +366,19 @@ function Te({ onSubmit: Te, latestReply: f, language: p, graceMs: Ee, onSupersed
 		J.fallbackReason,
 		J.privacyNotice,
 		q.isSupported,
-		_.isBlocked,
+		g.isBlocked,
 		$,
-		Ke,
-		qe,
 		Je,
-		Ze,
-		ze,
 		Ye,
 		Xe,
+		$e,
+		Ve,
+		Ze,
+		Qe,
 		U,
-		Fe,
-		Ie
+		Le,
+		Re
 	]);
 }
 //#endregion
-export { Te as useVoiceSession };
+export { De as useVoiceSession };

package/dist/shared/assistant/voice/handsfree/handsFreeController.d.ts

@@ -33,6 +33,9 @@ export interface ControllerDeps {
     onFallback: (reason: FallbackReason) => void;
     /** A barge-in that produced no words within `NOTHING_HEARD_MS`. */
     onNothingHeard?: () => void;
+    /** Timing only: end of the latest echo-approved voiced capture frame. */
+    onSpeechFrame?: (at: number) => void;
+    now?: () => number;
     isEnabled: () => boolean;
     prerollMs?: number;
     setTimer?: (fn: () => void, ms: number) => ReturnType<typeof setTimeout>;

package/dist/shared/assistant/voice/handsfree/handsFreeController.js

@@ -66,7 +66,11 @@ function u(u) {
 			rmsDb: e.rmsDb,
 			zcr: e.zcr
 		}, t.pass);
-		r.event === "onset" ? M(t.reason) : b && r.inSegment && u.transport.pushFrame(n), r.event === "end" && j();
+		if (r.event === "onset" && M(t.reason), b && r.voiced && t.pass) {
+			let t = u.now?.() ?? performance.now(), n = _?.context ? Math.max(0, (_.context.currentTime - e.ctxTime) * 1e3) : 0;
+			u.onSpeechFrame?.(t - n);
+		}
+		r.event !== "onset" && b && r.inSegment && u.transport.pushFrame(n), r.event === "end" && j();
 	}, P = () => {
 		if (!v && !_) return;
 		v = !1, k(), b && u.transport.cancelUtterance(b.id), b = null, g.invalidate(), u.transport.stop(), h.clear(), p.reset(), m.reset();

package/dist/shared/assistant/voice/handsfree/useHandsFreeInterrupt.d.ts

@@ -24,6 +24,7 @@ export interface UseHandsFreeInterruptOptions {
     language: string;
     /** The user's voice started while the assistant was audible. */
     onBargeIn: () => void;
+    onSpeechFrame?: (at: number) => void;
     onTranscript: (text: string, final: boolean, language: string | undefined) => void;
     /**
      * The backend stream's privacy-notice version this person accepted (it is

package/dist/shared/assistant/voice/handsfree/useHandsFreeInterrupt.js

@@ -57,6 +57,7 @@ function b(s) {
 			onTranscript: (e, t, n) => {
 				D(!1), C.current.onTranscript(e, t, n);
 			},
+			onSpeechFrame: (e) => C.current.onSpeechFrame?.(e),
 			onFallback: T,
 			onNothingHeard: () => D(!0),
 			isEnabled: () => C.current.enabled

package/dist/shared/assistant/voice/useVoiceInput.d.ts

@@ -94,6 +94,8 @@ export interface UseVoiceInputOptions {
      * here) means the user can freely edit while dictation continues.
      */
     onFinalize: (text: string) => void;
+    /** Browser endpoint notification time; not an acoustic timestamp. */
+    onSpeechEnd?: (at: number) => void;
     /**
      * BCP-47 tag to listen for. Omit to follow the app locale, which is what this
      * hook did unconditionally before (BOFF-7108).
@@ -137,4 +139,4 @@ export interface UseVoiceInputResult {
     toggle: () => void;
     clearError: () => void;
 }
-export declare function useVoiceInput({ onFinalize, language: requested, }: UseVoiceInputOptions): UseVoiceInputResult;
+export declare function useVoiceInput({ onFinalize, onSpeechEnd, language: requested, }: UseVoiceInputOptions): UseVoiceInputResult;