natural voice

frontend/src/lib/api.ts

@@ -74,6 +74,7 @@ export interface VoicePreview {
   rate: number;
   pitch: number;
   timbre: "masculine" | "feminine" | "any";
+  voice: string;
 }
 
 export interface PlatformInfo {
@@ -102,6 +103,7 @@ export interface AdminSettings {
   recoveryLimitMax: number;
   recoveryLimitWindowMin: number;
   voicePreviews: VoicePreview[];
+  ttsEnabled: boolean;
 }
 
 export interface AdminSettingsPatch {
@@ -115,6 +117,7 @@ export interface AdminSettingsPatch {
   recoveryLimitMax?: number;
   recoveryLimitWindowMin?: number;
   voicePreviews?: VoicePreview[];
+  ttsEnabled?: boolean;
 }
 
 export interface OllamaNode {
@@ -442,6 +445,27 @@ export const api = {
 
   tools: () => req<{ tools: { name: string; description: string; available: boolean; reason: string | null }[]; reportsEnabled: boolean; usernameChangeEnabled: boolean; voicePreviews: VoicePreview[] }>("/api/tools"),
 
+  /** Neural TTS provider status (null = browser speech fallback). */
+  ttsInfo: () => req<{ provider: "elevenlabs" | "openai" | "piper" | null; voices: string[]; available: boolean; reason: string | null }>("/api/tools/tts/info"),
+
+  /** Synthesize text server-side (neural voice). Returns audio bytes. */
+  ttsSpeak: async (text: string, voice?: string): Promise<Blob> => {
+    const res = await fetch("/api/tools/tts", {
+      method: "POST",
+      headers: { "Content-Type": "application/json", Authorization: `Bearer ${getToken()}` },
+      body: JSON.stringify({ text, ...(voice ? { voice } : {}) }),
+    });
+    if (res.status === 401) {
+      clearToken();
+      window.location.reload();
+      throw new ApiError(401, "Session expired.");
+    }
+    if (!res.ok) {
+      const data = (await res.json().catch(() => ({}))) as { error?: string };
+      throw new ApiError(res.status, data.error || `TTS failed (${res.status})`);
+    }
+    return await res.blob();
+  },
   /** Send an image to the platform OCR tool. Returns editable text (never the raw image). */
   ocrImage: async (file: File): Promise<{ text: string; truncated: boolean; chars: number }> => {
     const res = await fetch("/api/tools/ocr", {

frontend/src/lib/i18n.ts

@@ -605,6 +605,22 @@ const en = {
   "vfeat.timbre": "Timbre",
   "vfeat.rate": "Speed",
   "vfeat.pitch": "Pitch",
+  "vfeat.voice": "Server voice",
+  "vfeat.voiceHint": "ElevenLabs id / OpenAI name, empty = default",
+  "vfeat.voicePh": "auto",
+  "center.ttsTitle": "Neural voice playback",
+  "center.ttsDesc": "Server text-to-speech for message playback (ElevenLabs / OpenAI / piper). Off = browser speech only.",
+  "voice.engine": "Playback engine",
+  "voice.engineAuto": "Auto (neural if available)",
+  "voice.engineBrowser": "Browser speech",
+  "voice.engineServer": "Neural server voice",
+  "voice.engineHintOn": "Neural TTS via {provider}",
+  "voice.engineHintOff": "No neural provider configured — browser speech only",
+  "voice.serverVoice": "Neural voice",
+  "voice.serverVoiceHint": "ElevenLabs voice id",
+  "voice.piperNote": "Local piper voice (single configured model).",
+  "voice.noServer": "Configure ELEVENLABS_API_KEY, OPENAI_API_KEY or piper to unlock neural voices.",
+
 
 
   "msg.copy": "Copy message",
@@ -1297,6 +1313,22 @@ const es: Dict = {
   "vfeat.timbre": "Timbre",
   "vfeat.rate": "Speed",
   "vfeat.pitch": "Pitch",
+  "vfeat.voice": "Server voice",
+  "vfeat.voiceHint": "ElevenLabs id / OpenAI name, empty = default",
+  "vfeat.voicePh": "auto",
+  "center.ttsTitle": "Neural voice playback",
+  "center.ttsDesc": "Server text-to-speech for message playback (ElevenLabs / OpenAI / piper). Off = browser speech only.",
+  "voice.engine": "Playback engine",
+  "voice.engineAuto": "Auto (neural if available)",
+  "voice.engineBrowser": "Browser speech",
+  "voice.engineServer": "Neural server voice",
+  "voice.engineHintOn": "Neural TTS via {provider}",
+  "voice.engineHintOff": "No neural provider configured — browser speech only",
+  "voice.serverVoice": "Neural voice",
+  "voice.serverVoiceHint": "ElevenLabs voice id",
+  "voice.piperNote": "Local piper voice (single configured model).",
+  "voice.noServer": "Configure ELEVENLABS_API_KEY, OPENAI_API_KEY or piper to unlock neural voices.",
+
 
 
   "msg.copy": "Copiar mensaje",
@@ -1984,6 +2016,22 @@ const fr: Dict = {
   "vfeat.timbre": "Timbre",
   "vfeat.rate": "Speed",
   "vfeat.pitch": "Pitch",
+  "vfeat.voice": "Server voice",
+  "vfeat.voiceHint": "ElevenLabs id / OpenAI name, empty = default",
+  "vfeat.voicePh": "auto",
+  "center.ttsTitle": "Neural voice playback",
+  "center.ttsDesc": "Server text-to-speech for message playback (ElevenLabs / OpenAI / piper). Off = browser speech only.",
+  "voice.engine": "Playback engine",
+  "voice.engineAuto": "Auto (neural if available)",
+  "voice.engineBrowser": "Browser speech",
+  "voice.engineServer": "Neural server voice",
+  "voice.engineHintOn": "Neural TTS via {provider}",
+  "voice.engineHintOff": "No neural provider configured — browser speech only",
+  "voice.serverVoice": "Neural voice",
+  "voice.serverVoiceHint": "ElevenLabs voice id",
+  "voice.piperNote": "Local piper voice (single configured model).",
+  "voice.noServer": "Configure ELEVENLABS_API_KEY, OPENAI_API_KEY or piper to unlock neural voices.",
+
 
 
   "msg.copy": "Copier le message",
@@ -2659,6 +2707,22 @@ const ru: Dict = {
   "vfeat.timbre": "Timbre",
   "vfeat.rate": "Speed",
   "vfeat.pitch": "Pitch",
+  "vfeat.voice": "Server voice",
+  "vfeat.voiceHint": "ElevenLabs id / OpenAI name, empty = default",
+  "vfeat.voicePh": "auto",
+  "center.ttsTitle": "Neural voice playback",
+  "center.ttsDesc": "Server text-to-speech for message playback (ElevenLabs / OpenAI / piper). Off = browser speech only.",
+  "voice.engine": "Playback engine",
+  "voice.engineAuto": "Auto (neural if available)",
+  "voice.engineBrowser": "Browser speech",
+  "voice.engineServer": "Neural server voice",
+  "voice.engineHintOn": "Neural TTS via {provider}",
+  "voice.engineHintOff": "No neural provider configured — browser speech only",
+  "voice.serverVoice": "Neural voice",
+  "voice.serverVoiceHint": "ElevenLabs voice id",
+  "voice.piperNote": "Local piper voice (single configured model).",
+  "voice.noServer": "Configure ELEVENLABS_API_KEY, OPENAI_API_KEY or piper to unlock neural voices.",
+
 
 
   "msg.copy": "Скопировать сообщение",
@@ -3339,6 +3403,22 @@ const it: Dict = {
   "vfeat.timbre": "Timbre",
   "vfeat.rate": "Speed",
   "vfeat.pitch": "Pitch",
+  "vfeat.voice": "Server voice",
+  "vfeat.voiceHint": "ElevenLabs id / OpenAI name, empty = default",
+  "vfeat.voicePh": "auto",
+  "center.ttsTitle": "Neural voice playback",
+  "center.ttsDesc": "Server text-to-speech for message playback (ElevenLabs / OpenAI / piper). Off = browser speech only.",
+  "voice.engine": "Playback engine",
+  "voice.engineAuto": "Auto (neural if available)",
+  "voice.engineBrowser": "Browser speech",
+  "voice.engineServer": "Neural server voice",
+  "voice.engineHintOn": "Neural TTS via {provider}",
+  "voice.engineHintOff": "No neural provider configured — browser speech only",
+  "voice.serverVoice": "Neural voice",
+  "voice.serverVoiceHint": "ElevenLabs voice id",
+  "voice.piperNote": "Local piper voice (single configured model).",
+  "voice.noServer": "Configure ELEVENLABS_API_KEY, OPENAI_API_KEY or piper to unlock neural voices.",
+
 
 
   "msg.copy": "Copia messaggio",
@@ -4010,6 +4090,22 @@ const pt: Dict = {
   "vfeat.timbre": "Timbre",
   "vfeat.rate": "Speed",
   "vfeat.pitch": "Pitch",
+  "vfeat.voice": "Server voice",
+  "vfeat.voiceHint": "ElevenLabs id / OpenAI name, empty = default",
+  "vfeat.voicePh": "auto",
+  "center.ttsTitle": "Neural voice playback",
+  "center.ttsDesc": "Server text-to-speech for message playback (ElevenLabs / OpenAI / piper). Off = browser speech only.",
+  "voice.engine": "Playback engine",
+  "voice.engineAuto": "Auto (neural if available)",
+  "voice.engineBrowser": "Browser speech",
+  "voice.engineServer": "Neural server voice",
+  "voice.engineHintOn": "Neural TTS via {provider}",
+  "voice.engineHintOff": "No neural provider configured — browser speech only",
+  "voice.serverVoice": "Neural voice",
+  "voice.serverVoiceHint": "ElevenLabs voice id",
+  "voice.piperNote": "Local piper voice (single configured model).",
+  "voice.noServer": "Configure ELEVENLABS_API_KEY, OPENAI_API_KEY or piper to unlock neural voices.",
+
 
 
   "msg.copy": "Copiar mensagem",

frontend/src/lib/voice.ts

@@ -15,21 +15,28 @@ export interface VoicePrefs {
   pitch: number;
   /** "auto" detects per message, otherwise forces a BCP-47 prefix. */
   lang: string;
+  /** Playback engine: neural server TTS when available, else browser. */
+  engine: "auto" | "browser" | "server";
+  /** Server neural voice id (ElevenLabs voice_id / OpenAI voice name). */
+  serverVoice: string | null;
 }
 
 const PREFS_KEY = "ka:voice";
 
-export const DEFAULT_VOICE_PREFS: VoicePrefs = { voiceURI: null, rate: 1, pitch: 1, lang: "auto" };
+export const DEFAULT_VOICE_PREFS: VoicePrefs = { voiceURI: null, rate: 1, pitch: 1, lang: "auto", engine: "auto", serverVoice: null };
 
 export function getVoicePrefs(): VoicePrefs {
   try {
     const raw = JSON.parse(localStorage.getItem(PREFS_KEY) ?? "null") as Partial<VoicePrefs> | null;
     if (!raw || typeof raw !== "object") return { ...DEFAULT_VOICE_PREFS };
+    const engine = raw.engine === "browser" || raw.engine === "server" ? raw.engine : "auto";
     return {
       voiceURI: typeof raw.voiceURI === "string" ? raw.voiceURI : null,
       rate: typeof raw.rate === "number" && raw.rate >= 0.5 && raw.rate <= 2 ? raw.rate : 1,
       pitch: typeof raw.pitch === "number" && raw.pitch >= 0.5 && raw.pitch <= 2 ? raw.pitch : 1,
       lang: typeof raw.lang === "string" && raw.lang.length <= 12 ? raw.lang : "auto",
+      engine,
+      serverVoice: typeof raw.serverVoice === "string" && raw.serverVoice.length > 0 ? raw.serverVoice : null,
     };
   } catch {
     return { ...DEFAULT_VOICE_PREFS };
@@ -83,13 +90,6 @@ export interface SpeakOpts {
 
 export type VoiceTimbre = "masculine" | "feminine" | "any";
 
-export interface PreviewVoice {
-  lang: string;
-  rate: number;
-  pitch: number;
-  timbre: VoiceTimbre;
-}
-
 // Name fragments hinting a masculine / feminine voice (Chrome, Edge,
 // Safari and Android voices carry such markers; absence falls back to
 // any same-language voice — never an error).
@@ -158,43 +158,163 @@ export function stopSpeak(): void {
   }
 }
 
-/**
- * Play a preview line with ITS OWN voice character (timbre + rate + pitch),
- * never the user's global playback prefs — previews must sound different
- * from each other. Returns the resolved voice (null when unsupported) so
- * callers can offer "use this voice" adoption.
- */
-export function speakPreview(
-  text: string,
-  preview: PreviewVoice,
-  voices: SpeechSynthesisVoice[],
-  opts: SpeakOpts = {},
-): SpeechSynthesisVoice | null {
-  if (!ttsSupported()) return null;
-  const synth = window.speechSynthesis;
+// --- unified playback (neural server audio with browser fallback) ---
+
+export interface PlayHandle {
+  /** Engine that ended up playing (may differ from the request in auto). */
+  engine: "server" | "browser";
+  stop: () => void;
+  done: Promise<void>;
+}
+
+let currentAudio: HTMLAudioElement | null = null;
+
+function stopCurrentAudio(): void {
+  if (currentAudio) {
+    try {
+      currentAudio.pause();
+      const url = currentAudio.src;
+      currentAudio.removeAttribute("src");
+      currentAudio.load();
+      if (url.startsWith("blob:")) URL.revokeObjectURL(url);
+    } catch {
+      // ignore
+    }
+    currentAudio = null;
+  }
   try {
-    synth.cancel();
+    window.speechSynthesis?.cancel();
   } catch {
     // ignore
   }
-  const lang = preview.lang === "auto" ? detectLang(text) : preview.lang;
-  const utter = new SpeechSynthesisUtterance(text.slice(0, 2000));
-  utter.lang = lang;
-  utter.rate = preview.rate;
-  utter.pitch = preview.pitch;
-  const pick = pickVoice(voices, { lang, timbre: preview.timbre });
-  if (pick) utter.voice = pick;
-  utter.onend = () => opts.onend?.();
-  utter.onerror = () => opts.onerror?.("error");
-  try {
-    synth.speak(utter);
-  } catch {
-    return null;
-  }
-  return pick;
 }
 
-// --- speech recognition (STT) ---
+/** Stop whatever is playing (server audio or browser speech). */
+export function stopPlayback(): void {
+  stopCurrentAudio();
+}
+
+export interface PlayRequest {
+  text: string;
+  /** "server" forces neural, "browser" forces synthesis, "auto" tries neural first. */
+  engine: "auto" | "browser" | "server";
+  browser: {
+    voices: SpeechSynthesisVoice[];
+    /** "auto" detects per text, otherwise a BCP-47 prefix. */
+    lang: string;
+    rate: number;
+    pitch: number;
+    /** Explicit voice URI (user pref); otherwise timbre-matched. */
+    voiceURI?: string | null;
+    /** Preview character timbre (ignored when voiceURI is set). */
+    timbre?: VoiceTimbre;
+  };
+  /** Server neural voice id (preview line voice or user serverVoice). */
+  serverVoice?: string | null;
+  /** True when the platform TTS tool is available. */
+  serverAvailable: boolean;
+  fetchAudio: (text: string, voice?: string) => Promise<Blob>;
+  onstart?: () => void;
+  onend?: () => void;
+  onerror?: (msg: string) => void;
+}
+
+/**
+ * Play one text with the best engine: neural server audio when requested
+ * and available (shared <audio> element, user speed), otherwise the
+ * browser character voice. Single global playback — a new play stops the
+ * previous one. `done` resolves when playback fully ends (fallback
+ * included); it never rejects (errors go to onend/onerror like speech).
+ */
+export function playText(req: PlayRequest): PlayHandle {
+  stopPlayback();
+  const wantServer = req.serverAvailable && req.engine !== "browser";
+  if (wantServer) return playServer(req, req.engine === "auto");
+  return playBrowser(req);
+}
+
+function playServer(req: PlayRequest, allowFallback: boolean): PlayHandle {
+  let settled = false;
+  const stop = () => {
+    settled = true;
+    stopCurrentAudio();
+  };
+  const done = (async () => {
+    try {
+      const blob = await req.fetchAudio(req.text.slice(0, 2000), req.serverVoice ?? undefined);
+      if (settled) return;
+      await playAudioBlob(blob, req, () => settled);
+    } catch (e) {
+      if (settled) return;
+      if (allowFallback) {
+        await playBrowserAudio(req, () => settled);
+        return;
+      }
+      req.onerror?.(e instanceof Error ? e.message : "playback failed");
+    }
+  })();
+  return { engine: "server", stop, done };
+}
+
+function playAudioBlob(blob: Blob, req: PlayRequest, cancelled: () => boolean): Promise<void> {
+  return new Promise((resolve) => {
+    const url = URL.createObjectURL(blob);
+    const audio = new Audio(url);
+    audio.playbackRate = req.browser.rate;
+    currentAudio = audio;
+    const finish = (ok: boolean) => {
+      if (currentAudio === audio) stopCurrentAudio();
+      else {
+        try {
+          URL.revokeObjectURL(url);
+        } catch {
+          // ignore
+        }
+      }
+      if (!cancelled() && ok) req.onend?.();
+      resolve();
+    };
+    audio.onended = () => finish(true);
+    audio.onerror = () => finish(false);
+    req.onstart?.();
+    audio.play().catch(() => finish(false));
+  });
+}
+
+function playBrowser(req: PlayRequest): PlayHandle {
+  let settled = false;
+  const stop = () => {
+    settled = true;
+    stopSpeak();
+  };
+  const done = playBrowserAudio(req, () => settled);
+  return { engine: "browser", stop, done };
+}
+
+function playBrowserAudio(req: PlayRequest, cancelled: () => boolean): Promise<void> {
+  return new Promise((resolve) => {
+    const b = req.browser;
+    const lang = b.lang === "auto" ? detectLang(req.text) : b.lang;
+    const explicit = b.voiceURI ? b.voices.find((v) => v.voiceURI === b.voiceURI) : undefined;
+    const pick = explicit ?? pickVoice(b.voices, { lang, timbre: b.timbre ?? "any" });
+    const finish = (ok: boolean, msg?: string) => {
+      if (cancelled()) {
+        resolve();
+        return;
+      }
+      if (ok) req.onend?.();
+      else req.onerror?.(msg ?? "error");
+      resolve();
+    };
+    const ok = speak(
+      req.text,
+      { voiceURI: pick?.voiceURI ?? null, rate: b.rate, pitch: b.pitch, lang, engine: "browser", serverVoice: null },
+      b.voices,
+      { onstart: () => req.onstart?.(), onend: () => finish(true), onerror: (m) => finish(false, m) },
+    );
+    if (!ok) finish(false, "unsupported");
+  });
+}
 
 export interface Recognizer {
   lang: string;