feat: voix mains-libres + wake-word CHLOVA, fin Phase 6 v1 (v0.29.0)

Mode mains-libres : écoute en boucle déclenchée par le wake-word
« CHLOVA … » (extractCommand), micro en pause pendant le TTS pour éviter
l'auto-écoute ; réponses lues d'office. Bouton Libre + indicateur. 100%
navigateur. Build OK.

Palier de risque : reversible (front).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Kantin-Petit
2026-06-23 07:24:41 +02:00
parent 76ad3b62fd
commit 476c89ce3d
4 changed files with 127 additions and 14 deletions
+11 -2
View File
@@ -23,5 +23,14 @@ Le backend doit tourner avec l'auth configurée (`CHLOVA_ADMIN_*`, voir
| `src/pages/Chat.tsx` | Conversation agent (v0.23.0). |
| `src/pages/Review.tsx` | Need-review : approuver/refuser (v0.24.0). |
## Périmètre v1
Login → Chat → Review. Voix + app RN : phases ultérieures (API commune réutilisée).
## Voix (Phase 6)
100 % navigateur (Web Speech API), zéro backend/GPU :
- **Parler** : dictée push-to-talk (fr-FR) → envoyée à l'agent.
- **Voix ON/OFF** : lecture vocale des réponses (TTS), réglage persistant.
- **Libre** : mains-libres, déclenché par le wake-word « CHLOVA … » ; le micro se
met en pause pendant la synthèse pour éviter l'auto-écoute.
STT = Chrome/Edge (webkit) ; TTS = large support. Dégrade proprement sinon.
## Périmètre
Login → Chat (+ voix) → Review. App RN : phase ultérieure (API commune réutilisée).
+22 -3
View File
@@ -42,7 +42,7 @@ export function Chat() {
try {
const { reply } = await api.chat(token, t);
setMessages((m) => [...m, { role: "assistant", text: reply }]);
if (speakReplies) speech.speak(reply);
if (speakReplies || speech.handsFree) speech.speak(reply);
} catch (err) {
if (err instanceof ApiError && err.status === 401) {
logout();
@@ -62,13 +62,18 @@ export function Chat() {
};
const mic = (): void => {
if (speech.listening) {
if (speech.listening && !speech.handsFree) {
speech.stopListening();
return;
}
speech.listen((text) => void sendText(text)); // dicter → envoyer
};
const toggleHandsFree = (): void => {
if (speech.handsFree) speech.stopHandsFree();
else speech.startHandsFree((text) => void sendText(text)); // « CHLOVA … » → envoyer
};
return (
<div className="flex h-full flex-col">
<div className="flex-1 overflow-y-auto px-4 py-4 space-y-3">
@@ -86,6 +91,9 @@ export function Chat() {
</div>
))}
{busy && <p className="text-muted text-sm animate-pulse">CHLOVA réfléchit</p>}
{speech.handsFree && !busy && !speech.speaking && (
<p className="text-accent text-sm">Mains libres dis « CHLOVA »</p>
)}
{speech.speaking && <p className="text-accent text-sm">Lecture vocale</p>}
{error && <p role="alert" className="text-danger text-sm">{error}</p>}
<div ref={bottom} />
@@ -110,7 +118,7 @@ export function Chat() {
onChange={(e) => setInput(e.target.value)}
disabled={busy}
/>
{speech.sttSupported && (
{speech.sttSupported && !speech.handsFree && (
<button
type="button"
onClick={mic}
@@ -120,6 +128,17 @@ export function Chat() {
{speech.listening ? "Stop" : "Parler"}
</button>
)}
{speech.sttSupported && (
<button
type="button"
onClick={toggleHandsFree}
aria-label={speech.handsFree ? "Couper les mains libres" : "Activer les mains libres"}
title="Mains libres (wake-word CHLOVA)"
className={`rounded-md border px-3 py-2 text-sm cursor-pointer ring-accent ${speech.handsFree ? "border-accent text-accent" : "border-border text-muted"}`}
>
{speech.handsFree ? "Libre ON" : "Libre"}
</button>
)}
<button
type="submit"
disabled={busy || !input.trim()}
+87 -9
View File
@@ -4,9 +4,10 @@ import { useCallback, useEffect, useRef, useState } from "react";
* Voix (Phase 6) — 100 % navigateur (Web Speech API), aucun backend ni GPU.
* - STT : SpeechRecognition (webkit) — Chrome/Edge. Dégrade proprement ailleurs.
* - TTS : speechSynthesis — large support.
* - Mains-libres : écoute en boucle, déclenchée par le wake-word « CHLOVA ».
* Met le micro en pause pendant la synthèse vocale (évite l'auto-écoute).
*/
// Typages minimaux (Web Speech API non standardisée dans lib.dom).
interface SREvent {
results: ArrayLike<ArrayLike<{ transcript: string }>>;
}
@@ -40,23 +41,41 @@ function makeRecognition(): SR | null {
return r;
}
const WAKE = /\b(chlova|clova|klova)\b[\s,:.]*/i;
/** Renvoie la commande après le wake-word, ou null si absent. */
export function extractCommand(transcript: string): string | null {
const m = WAKE.exec(transcript);
if (!m) return null;
const cmd = transcript.slice(m.index + m[0].length).trim();
return cmd.length > 0 ? cmd : null;
}
export interface UseSpeech {
sttSupported: boolean;
ttsSupported: boolean;
listening: boolean;
speaking: boolean;
handsFree: boolean;
listen: (onText: (text: string) => void) => void;
stopListening: () => void;
speak: (text: string) => void;
cancelSpeak: () => void;
startHandsFree: (onCommand: (text: string) => void) => void;
stopHandsFree: () => void;
}
export function useSpeech(): UseSpeech {
const recRef = useRef<SR | null>(null);
const handsFreeRef = useRef(false);
const onCommandRef = useRef<(t: string) => void>(() => {});
const armRef = useRef<() => void>(() => {});
const [listening, setListening] = useState(false);
const [speaking, setSpeaking] = useState(false);
const [handsFree, setHandsFree] = useState(false);
const sttSupported = typeof window !== "undefined" && !!(window.SpeechRecognition ?? window.webkitSpeechRecognition);
const sttSupported =
typeof window !== "undefined" && !!(window.SpeechRecognition ?? window.webkitSpeechRecognition);
const ttsSupported = typeof window !== "undefined" && "speechSynthesis" in window;
const stopListening = useCallback((): void => {
@@ -89,18 +108,77 @@ export function useSpeech(): UseSpeech {
window.speechSynthesis.cancel();
const u = new SpeechSynthesisUtterance(text);
u.lang = "fr-FR";
u.onend = (): void => setSpeaking(false);
u.onerror = (): void => setSpeaking(false);
const done = (): void => {
setSpeaking(false);
// En mains-libres, on réarme l'écoute APRÈS avoir parlé (anti auto-écoute).
if (handsFreeRef.current) armRef.current();
};
u.onend = done;
u.onerror = done;
setSpeaking(true);
window.speechSynthesis.speak(u);
},
[ttsSupported],
);
useEffect(() => () => {
recRef.current?.abort();
if (ttsSupported) window.speechSynthesis.cancel();
}, [ttsSupported]);
// Boucle mains-libres : une écoute → détecte le wake-word → commande → réarme.
armRef.current = (): void => {
if (!handsFreeRef.current) return;
if (ttsSupported && window.speechSynthesis.speaking) return; // attend la fin du TTS
const r = makeRecognition();
if (!r) return;
recRef.current = r;
r.onresult = (e): void => {
const text = e.results?.[0]?.[0]?.transcript ?? "";
const cmd = extractCommand(text);
if (cmd) onCommandRef.current(cmd);
};
const rearm = (): void => {
setListening(false);
if (handsFreeRef.current && !(ttsSupported && window.speechSynthesis.speaking)) {
setTimeout(() => armRef.current(), 400);
}
};
r.onend = rearm;
r.onerror = rearm;
setListening(true);
r.start();
};
return { sttSupported, ttsSupported, listening, speaking, listen, stopListening, speak, cancelSpeak };
const startHandsFree = useCallback((onCommand: (text: string) => void): void => {
onCommandRef.current = onCommand;
handsFreeRef.current = true;
setHandsFree(true);
armRef.current();
}, []);
const stopHandsFree = useCallback((): void => {
handsFreeRef.current = false;
setHandsFree(false);
recRef.current?.abort();
setListening(false);
}, []);
useEffect(
() => () => {
handsFreeRef.current = false;
recRef.current?.abort();
if (ttsSupported) window.speechSynthesis.cancel();
},
[ttsSupported],
);
return {
sttSupported,
ttsSupported,
listening,
speaking,
handsFree,
listen,
stopListening,
speak,
cancelSpeak,
startHandsFree,
stopHandsFree,
};
}