akari-video 0.1.82 → 0.1.83

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "akari-video",
-  "version": "0.1.82",
+  "version": "0.1.83",
   "description": "AKARI Video launcher CLI — start an AI-edited video project from any directory: scaffold, connection check, then hand over to Claude Code (or opencode). AKARI Video を opencode や Claude Code で、どのディレクトリからでも始めるための `akari` ランチャー CLI。接続確認（doctor）→ 未セットアップならプロジェクト雛形を作成 → AI エージェントを起動する。外部 npm 依存ゼロ（Node.js 組み込みモジュールのみ）。",
   "type": "module",
   "bin": {

package/src/narration-command.mjs

@@ -3,11 +3,13 @@
 
 import fs from "node:fs";
 import os from "node:os";
+import crypto from "node:crypto";
 import path from "node:path";
 import { spawn, spawnSync } from "node:child_process";
 import { fileURLToPath, pathToFileURL } from "node:url";
 import { resolveLauncherAssets } from "./repo-assets.mjs";
-import { listProfiles, readFalKey, resolveVoiceProfile } from "./voice-command.mjs";
+import { listProfiles, readFalKey, readProviderKey, resolveVoiceProfile } from "./voice-command.mjs";
+import { FAL_TTS_ENGINES, DIRECT_TTS_ENGINES, falTtsEngine, ttsEngine, estimateTtsCost, referenceDataUri } from "./tts-engines.mjs";
 
 const VOICEVOX_BASE_URL = "http://127.0.0.1:50021";
 const VOICEVOX_RUN_ENV = "VOICEVOX_RUN";
@@ -38,22 +40,12 @@ function irodoriTimeout(env = process.env) {
   if (!Number.isSafeInteger(value) || value <= 0) throw new PublicError("AKARI_IRODORI_TIMEOUT_MS は正の整数にしてください", 2);
   return value;
 }
-const GEMINI_VOICES = Object.entries({
-  Leda: "Youthful", Achernar: "Soft", Achird: "Friendly", Algenib: "Gravelly",
-  Algieba: "Smooth", Alnilam: "Firm", Aoede: "Breezy", Autonoe: "Bright",
-  Callirrhoe: "Easy-going", Charon: "Informative", Despina: "Smooth",
-  Enceladus: "Breathy", Erinome: "Clear", Fenrir: "Excitable", Gacrux: "Mature",
-  Iapetus: "Clear", Kore: "Firm", Laomedeia: "Upbeat", Orus: "Firm",
-  Pulcherrima: "Forward", Puck: "Upbeat", Rasalgethi: "Informative",
-  Sadachbia: "Lively", Sadaltager: "Knowledgeable", Schedar: "Even",
-  Sulafat: "Warm", Umbriel: "Easy-going", Vindemiatrix: "Gentle",
-  Zephyr: "Bright", Zubenelgenubi: "Casual",
-}).map(([id, description]) => ({ id, label: `${id}（${description}）`, ...(id === "Leda" ? { default: true } : {}) }));
+const GEMINI_VOICES = falTtsEngine('gemini-tts').voices;
 
 const usage = [
   "使い方:",
   "  akari narration generate \\",
-  "    --project <projectDir> --engine <voicevox|gemini-tts|irodori|fal-qwen3> \\",
+  "    --project <projectDir> --engine <voicevox|gemini-tts|irodori|fal-qwen3|gemini-3.1-flash-tts|elevenlabs-v3|minimax-2.6-hd|chatterbox|index-tts-2|fish-s2.1-pro|gemini-3.8-flash-tts> \\",
   "    (--reading-file <読み原稿.txt> | --text <原稿>) [--script-file <表示原稿.txt>] \\",
   "    [--t <タイムライン秒>] [--gain-db 0] [--id n-0001] \\",
   "    [--speaker 3] [--voice Leda] [--style <text>] [--speed 1] \\",
@@ -305,8 +297,8 @@ function parseArguments(argv) {
   }
 
   if (!options.project) throw new PublicError("--project が必要です");
-  if (!["voicevox", "gemini-tts", "irodori", "fal-qwen3"].includes(options.engine)) {
-    throw new PublicError("--engine には voicevox、gemini-tts、irodori または fal-qwen3 を指定してください", 2);
+  if (!["voicevox", "irodori"].includes(options.engine) && !ttsEngine(options.engine)) {
+    throw new PublicError(`--engine には voicevox、irodori、${[...FAL_TTS_ENGINES, ...DIRECT_TTS_ENGINES].map(engine => engine.id).join('、')} を指定してください`, 2);
   }
   if (options.engine === "irodori") {
     options.voice ??= "narrator-male";
@@ -314,10 +306,11 @@ function parseArguments(argv) {
     if (options.voice === "custom" && !options.style?.trim()) throw new PublicError("自分で書く声には --style が必要です", 2);
     options.irodori = irodoriEndpoint(options.irodoriUrl);
     irodoriTimeout();
-  } else options.voice ??= "Leda";
+  } else options.voice ??= ttsEngine(options.engine)?.default_voice ?? (ttsEngine(options.engine) ? null : 'Leda');
   if (!options.readingFile && !options.text) throw new PublicError("--reading-file または --text が必要です");
-  if (options.engine === "gemini-tts" && !GEMINI_VOICES.some(({ id }) => id === options.voice)) {
-    throw new PublicError(`--voice に使える声: ${GEMINI_VOICES.map(({ id }) => id).join(", ")}`, 2);
+  const catalogVoices = ttsEngine(options.engine)?.voices;
+  if (catalogVoices && !options.profile && !catalogVoices.some(({ id }) => id === options.voice)) {
+    throw new PublicError(`--voice に使える声: ${catalogVoices.map(({ id }) => id).join(", ")}`, 2);
   }
   if (options.speed !== null && (!isFiniteNumber(options.speed) || options.speed < (options.engine === "irodori" ? 0.25 : 0.5) || options.speed > (options.engine === "irodori" ? 4 : 2))) {
     throw new PublicError(options.engine === "irodori" ? "--speed は 0.25 から 4.0 の範囲で指定してください" : "--speed は 0.5 から 2.0 の範囲で指定してください", 2);
@@ -335,8 +328,11 @@ function parseArguments(argv) {
   if (options.id !== null && !/^n-\d{4}$/.test(options.id)) {
     throw new PublicError("--id は n- に続く 4 桁の数字である必要があります（例: n-0001）");
   }
-  if (options.engine === "fal-qwen3" && !options.profile) {
-    throw new PublicError("--engine fal-qwen3 には --profile が必要です");
+  if ((options.engine === "fal-qwen3" || options.engine === "index-tts-2") && !options.profile) {
+    throw new PublicError(`--engine ${options.engine} には --profile が必要です`);
+  }
+  if (options.engine === 'elevenlabs-v3' && (options.text?.length ?? 0) > 5000) {
+    throw new PublicError('ElevenLabs v3 は 1 回 5000 字までです。原稿を短くしてください', 2);
   }
 
   options.project = path.resolve(options.project);
@@ -378,7 +374,7 @@ function computeNextId(projectDir) {
 }
 
 function extensionFor(engine) {
-  return ["voicevox", "irodori"].includes(engine) ? "wav" : "mp3";
+  return ["voicevox", "irodori", 'chatterbox', 'gemini-3.8-flash-tts'].includes(engine) ? "wav" : "mp3";
 }
 
 function relativeOutputPath(id, engine) {
@@ -406,6 +402,38 @@ function readProfileMeta(profileName) {
   return { ...meta, embedding_source_url };
 }
 
+function guardedProfile(profileName, engine) {
+  let record;
+  try { record = resolveVoiceProfile(profileName); }
+  catch { throw new PublicError(`声プロファイルが見つかりません: ${profileName}`, 2); }
+  const { meta } = record;
+  if (meta.consent?.self_voice !== true || meta.consent?.cloud_upload !== true ||
+      !(meta.reference?.verification?.score >= 0.7)) {
+    throw new PublicError('クラウド送信には本人同意・cloud_upload 同意・原稿照合 0.7 以上が必要です', 2);
+  }
+  if (engine.supports.clone === 'registered' &&
+      (engine.id === 'gemini-3.8-flash-tts' ?
+        (!/^voice_[A-Za-z0-9_-]+$/.test(meta.engines?.[engine.id]?.voice_id ?? '') || meta.engines[engine.id].stale) :
+        (!meta.engines?.[engine.id]?.custom_voice_id || meta.engines[engine.id].stale))) {
+    throw new PublicError('この声には指定した作り手の写しがありません。akari voice copy で作ってください', 2);
+  }
+  const name = meta.reference?.file;
+  if (engine.supports.clone === 'per-request' &&
+      (!name || path.basename(name) !== name || !fs.existsSync(path.join(record.dir, name)))) {
+    throw new PublicError('正本の録音が見つかりません', 2);
+  }
+  if (engine.supports.clone === 'per-request' && meta.reference.sha256 &&
+      crypto.createHash('sha256').update(fs.readFileSync(path.join(record.dir, name))).digest('hex') !== meta.reference.sha256) {
+    throw new PublicError('正本の録音が保存時から変わっています', 2);
+  }
+  return record;
+}
+
+function profileReferenceDataUri(record) {
+  try { return referenceDataUri(fs.readFileSync(path.join(record.dir, record.meta.reference.file))); }
+  catch (error) { throw new PublicError(error.message, 2); }
+}
+
 function irodoriProfileVoice(profileName) {
   let meta;
   try { meta = resolveVoiceProfile(profileName).meta; }
@@ -416,13 +444,7 @@ function irodoriProfileVoice(profileName) {
 }
 
 function buildFalPayload(readingText, meta) {
-  return {
-    text: readingText,
-    language: "Japanese",
-    speaker_voice_embedding_file_url: meta.embedding_source_url,
-    reference_text: meta.reference_text,
-    max_new_tokens: 2048,
-  };
+  return falTtsEngine('fal-qwen3').buildPayload({ text: readingText, profileMeta: meta });
 }
 
 function estimateFalCostUsd(readingText) {
@@ -432,53 +454,28 @@ function estimateGeminiTtsCostUsd(chars) {
   return Number(((chars / 1000) * GEMINI_USD_PER_1000_CHARS).toFixed(6));
 }
 
-async function synthesizeGeminiTts(readingText, { voice, model, style, falKey }) {
-  const payload = {
-    prompt: readingText, voice, model,
-    output_format: "mp3", language_code: "Japanese (Japan)",
-    ...(style ? { style_instructions: style } : {}),
-  };
+async function synthesizeCatalog(engine, payload, falKey) {
   let response;
   try {
-    response = await fetch(GEMINI_TTS_URL, {
-      method: "POST",
-      headers: { Authorization: `Key ${falKey}`, "Content-Type": "application/json" },
-      body: JSON.stringify(payload),
-    });
-  } catch {
-    throw new PublicError("fal API へのネットワーク接続に失敗しました");
-  }
+    const request = engine.request(payload, falKey);
+    response = await fetch(request.url, request.options);
+  } catch { throw new PublicError('fal API へのネットワーク接続に失敗しました'); }
   if (!response.ok) throw new PublicError(`fal API が HTTP ${response.status} を返しました`);
-  const audioUrl = (await response.json())?.audio?.url;
-  if (typeof audioUrl !== "string" || !audioUrl) throw new PublicError("fal API の応答に audio.url がありません");
+  const audioUrl = engine.parseAudio(await response.json());
+  if (!audioUrl) throw new PublicError('fal API の応答に audio.url がありません');
   const audioResponse = await fetch(audioUrl);
   if (!audioResponse.ok) throw new PublicError(`生成音声の取得に失敗しました（HTTP ${audioResponse.status}）`);
   return Buffer.from(await audioResponse.arrayBuffer());
 }
 
-async function synthesizeFal(readingText, meta, falKey) {
-  const payload = buildFalPayload(readingText, meta);
+async function synthesizeDirect(engine, payload, key) {
   let response;
-  try {
-    response = await fetch(FAL_TTS_URL, {
-      method: "POST",
-      headers: { Authorization: `Key ${falKey}`, "Content-Type": "application/json" },
-      body: JSON.stringify(payload),
-    });
-  } catch {
-    throw new PublicError("fal API へのネットワーク接続に失敗しました");
-  }
-  if (!response.ok) throw new PublicError(`fal API が HTTP ${response.status} を返しました`);
-  const result = await response.json();
-  const audioUrl = result?.audio?.url;
-  if (typeof audioUrl !== "string" || !audioUrl) {
-    throw new PublicError("fal API の応答に audio.url がありません");
-  }
-  const audioResponse = await fetch(audioUrl);
-  if (!audioResponse.ok) {
-    throw new PublicError(`生成音声の取得に失敗しました（HTTP ${audioResponse.status}）`);
-  }
-  return Buffer.from(await audioResponse.arrayBuffer());
+  try { const request = engine.request(payload, key); response = await fetch(request.url, request.options); }
+  catch { throw new PublicError(`${engine.label} API へのネットワーク接続に失敗しました`); }
+  if (!response.ok) throw new PublicError(`${engine.label} API が HTTP ${response.status} を返しました`);
+  if (engine.provider === 'fish-audio') return Buffer.from(await response.arrayBuffer());
+  try { return engine.parseAudio(await response.json()); }
+  catch (error) { throw new PublicError(error.message); }
 }
 
 // --- VOICEVOX ローカルエンジン ---
@@ -804,22 +801,41 @@ async function listEngines(runtime = {}) {
   const falAvailability = configured
     ? { state: "available", label: "fal を使用できます" }
     : { state: "unconfigured", label: "fal の鍵を登録" };
+  const directAvailability = Object.fromEntries([['fish-audio', 'FISH_AUDIO_API_KEY', 'Fish Audio'],
+    ['google-ai', 'GEMINI_API_KEY', 'Google AI']].map(([provider, name, label]) => {
+    let hasKey = false;
+    try { hasKey = Boolean(readProviderKey(name, runtime.env || process.env)); } catch { /* 一覧は維持する */ }
+    return [provider, hasKey ? { state: 'available', label: `${label} を使用できます` }
+      : { state: 'unconfigured', label: `${label} の鍵を登録` }];
+  }));
   const profilesWithFal = listFalProfiles(runtime).length;
   let endpoint;
   try { endpoint = irodoriEndpoint(runtime.irodoriUrl, runtime.env || process.env); }
   catch (error) { if (!(error instanceof PublicError)) throw error; }
   const irodoriAvailable = endpoint ? await probeIrodori(endpoint, runtime.fetchImpl || fetch) : false;
   return { version: 1, engines: [
-    { id: "voicevox", label: "VOICEVOX", place: "local", price: { usd_per_1000_chars: 0, verified: true }, availability: voicevox, credit_required: true, supports: { speed: true, style: false } },
-    { id: "gemini-tts", label: "Gemini 2.5 Flash TTS", place: "cloud", provider: "fal", price: { usd_per_1000_chars: GEMINI_USD_PER_1000_CHARS, verified: false, as_of: "2026-09-22" }, availability: falAvailability, default_voice: "Leda", credit_required: false, supports: { speed: false, style: true } },
+    { id: "voicevox", label: "VOICEVOX", place: "local", price: { usd_per_1000_chars: 0, verified: true }, availability: voicevox, credit_required: true, supports: { speed: true, style: false, clone: 'none' }, group: 'local' },
+    { id: "gemini-tts", label: "Gemini 2.5 Flash TTS", place: "cloud", provider: "fal", price: { usd_per_1000_chars: GEMINI_USD_PER_1000_CHARS, verified: false, as_of: "2026-09-22" }, availability: falAvailability, default_voice: "Leda", credit_required: false, supports: { speed: false, style: true, clone: 'none' }, group: 'cloud' },
     { id: "irodori", label: "彩（Irodori-TTS）", place: endpoint?.network ? "network" : "local", experimental: true,
       price: { usd_per_1000_chars: 0, verified: true }, credit_required: false, default_voice: "narrator-male",
-      supports: { speed: true, style: true }, availability: !endpoint
+      supports: { speed: true, style: true, clone: 'registered' }, group: 'local', availability: !endpoint
         ? { state: "unconfigured", label: "接続先 URL が正しくありません（お試し）", detail: { setup_url: IRODORI_SETUP_URL } }
         : irodoriAvailable
         ? { state: "available", label: "お試し · 接続済み", detail: { url: endpoint.server } }
         : { state: "unconfigured", label: "Irodori サーバーにつながりません（お試し）", detail: { setup_url: IRODORI_SETUP_URL } } },
-    { id: "fal-qwen3", label: "fal Qwen3-TTS", place: "cloud", provider: "fal", price: { usd_per_1000_chars: FAL_USD_PER_1000_CHARS, verified: false }, availability: !configured ? { ...falAvailability, detail: { profiles_with_fal: profilesWithFal } } : profilesWithFal ? { state: "available", label: "声プロファイルを使用できます", detail: { profiles_with_fal: profilesWithFal } } : { state: "needs", label: "fal の写しがある声がありません（自分の声をつくる）", detail: { profiles_with_fal: 0 } }, credit_required: false, supports: { speed: false, style: false } },
+    { id: "fal-qwen3", label: "fal Qwen3-TTS", place: "cloud", provider: "fal", price: { usd_per_1000_chars: FAL_USD_PER_1000_CHARS, verified: false }, availability: !configured ? { ...falAvailability, detail: { profiles_with_fal: profilesWithFal } } : profilesWithFal ? { state: "available", label: "声プロファイルを使用できます", detail: { profiles_with_fal: profilesWithFal } } : { state: "needs", label: "fal の写しがある声がありません（自分の声をつくる）", detail: { profiles_with_fal: 0 } }, credit_required: false, supports: { speed: false, style: false, clone: 'registered' }, group: 'cloud' },
+    ...FAL_TTS_ENGINES.filter(engine => !['gemini-tts', 'fal-qwen3'].includes(engine.id)).map(engine => ({
+      id: engine.id, label: engine.label, place: engine.place, provider: engine.provider,
+      group: 'cloud', endpoint: engine.endpoint, price: engine.price, ...(engine.caution ? { caution: engine.caution } : {}),
+      availability: falAvailability, default_voice: engine.default_voice, credit_required: false,
+      supports: engine.supports,
+    })),
+    ...DIRECT_TTS_ENGINES.map(engine => ({
+      id: engine.id, label: engine.label, place: engine.place, provider: engine.provider,
+      group: 'cloud', endpoint: engine.endpoint, price: engine.price,
+      availability: directAvailability[engine.provider], default_voice: engine.default_voice,
+      credit_required: false, supports: engine.supports,
+    })),
   ] };
 }
 
@@ -827,6 +843,13 @@ async function listVoices(engine, runtime = {}) {
   if (engine === "irodori") return [...IRODORI_RECIPES.map(({ id, label, default: isDefault }) => ({ id, label, ...(isDefault ? { default: true } : {}) })),
     { id: "custom", label: "自分で書く（声の指示）" }];
   if (engine === "gemini-tts") return GEMINI_VOICES;
+  const catalog = ttsEngine(engine);
+  if (catalog && !['fal-qwen3'].includes(engine)) {
+    if (catalog.voices) return catalog.voices;
+    return catalog.supports.clone === 'per-request' ? listProfiles(resolveProfileEnv(runtime))
+      .filter(profile => profile.consent.self_voice && profile.consent.cloud_upload && profile.verification?.score >= 0.7)
+      .map(profile => ({ id: profile.id, label: profile.label })) : [];
+  }
   if (engine === "voicevox") {
     const handle = await ensureVoicevoxEngine();
     try {
@@ -880,6 +903,25 @@ async function runDryRun(options, readingText, io) {
     }, io.log);
     return;
   }
+  if (DIRECT_TTS_ENGINES.some(engine => engine.id === options.engine)) {
+    const engine = ttsEngine(options.engine);
+    const record = options.profile ? guardedProfile(options.profile, engine) : null;
+    if (record && engine.supports.clone === 'per-request') {
+      profileReferenceDataUri(record); // 送信前と同じ容量ガード。録音は表示しない。
+      if (!record.meta.reference_text) throw new PublicError('声プロファイルの reference_text がありません', 2);
+    }
+    const body = engine.buildPayload({ text: readingText,
+      voice: engine.id === 'gemini-3.8-flash-tts' && record ? record.meta.engines[engine.id].voice_id : options.voice, style: options.style,
+      ...(record && engine.supports.clone === 'per-request' ? { referenceAudio: Buffer.from('<audio bytes>'), referenceText: record.meta.reference_text } : {}) });
+    if (record && engine.supports.clone === 'per-request') body.references[0].audio = '<audio bytes>';
+    emit({ dry_run: true, engine: engine.id, output_path: outputPath,
+      estimated_cost_usd: estimateTtsCost(engine, readingText),
+      request: { endpoint: engine.endpoint, headers: engine.provider === 'fish-audio'
+        ? { Authorization: `Bearer ${maskKey(readProviderKey('FISH_AUDIO_API_KEY'))}`, model: 's2.1-pro',
+          'Content-Type': record ? 'application/msgpack' : 'application/json' }
+        : { 'x-goog-api-key': maskKey(readProviderKey('GEMINI_API_KEY')), 'Content-Type': 'application/json' }, body } }, io.log);
+    return;
+  }
 
   if (options.engine === "gemini-tts") {
     emit({
@@ -898,6 +940,21 @@ async function runDryRun(options, readingText, io) {
         ...(profileVoice ? {} : { irodori: { caption: options.style?.trim() || IRODORI_RECIPES.find(recipe => recipe.id === options.voice)?.caption } }) } } }, io.log);
     return;
   }
+  if (!['fal-qwen3'].includes(options.engine)) {
+    const engine = ttsEngine(options.engine);
+    const record = options.profile && engine.supports.clone !== 'none' ? guardedProfile(options.profile, engine) : null;
+    let audioUrl;
+    if (record && engine.supports.clone === 'per-request') {
+      profileReferenceDataUri(record); // 容量を確認し、録音そのものは dry-run に表示しない。
+      audioUrl = '<data:audio/wav;base64,...>';
+    }
+    const payload = engine.buildPayload({ text: readingText, voice: options.voice, style: options.style,
+      speed: options.speed, profileMeta: record?.meta, audioUrl });
+    emit({ dry_run: true, engine: engine.id, output_path: outputPath,
+      estimated_cost_usd: estimateTtsCost(engine, readingText),
+      request: { endpoint: `https://fal.run/${engine.endpoint}`, body: payload } }, io.log);
+    return;
+  }
   const falKey = resolveFalKey();
   const meta = readProfileMeta(options.profile);
   const payload = buildFalPayload(readingText, meta);
@@ -916,7 +973,7 @@ async function runDryRun(options, readingText, io) {
 }
 
 function durationForAudio(buffer, outputPath, engine, warnings) {
-  if (["voicevox", "irodori"].includes(engine)) {
+  if (["voicevox", "irodori", 'chatterbox', 'gemini-3.8-flash-tts'].includes(engine)) {
     if (buffer.length >= 44 && buffer.toString("ascii", 0, 4) === "RIFF") {
       const bytesPerSecond = buffer.readUInt32LE(28);
       const dataOffset = buffer.indexOf("data", 36, "ascii");
@@ -939,6 +996,12 @@ async function runGenerate(options, io) {
   const readingText = options.readingFile ? readTextFile(options.readingFile, "読み原稿") : options.text.trim();
   const scriptText = options.scriptFile ? readTextFile(options.scriptFile, "表示原稿") : options.text;
   if (!readingText) throw new PublicError("読み原稿が空です", 2);
+  if (options.engine === 'elevenlabs-v3' && readingText.length > 5000) {
+    throw new PublicError('ElevenLabs v3 は 1 回 5000 字までです。原稿を短くしてください', 2);
+  }
+  if (options.engine === 'chatterbox' && readingText.length > 300) {
+    throw new PublicError('Chatterbox 多言語版は 1 回 300 字までです。原稿を短くしてください', 2);
+  }
 
   if (options.dryRun) {
     await runDryRun(options, readingText, io);
@@ -953,7 +1016,7 @@ async function runGenerate(options, io) {
   let provenance;
   let costUsd = 0;
   const warnings = [];
-  if (options.speed !== null && !["voicevox", "irodori"].includes(options.engine)) {
+  if (options.speed !== null && !["voicevox", "irodori", 'minimax-2.6-hd'].includes(options.engine)) {
     const warning = `${options.engine} は --speed に対応していないため無視しました`;
     warnings.push(warning);
     io.logError(warning);
@@ -986,28 +1049,60 @@ async function runGenerate(options, io) {
     provenance = { provider: "irodori", engine: "irodori-tts-v4-small", voice: options.profile ? `profile:${options.profile}` : options.style?.trim() ? "caption:custom" : `recipe:${options.voice}`,
       generated_at: new Date().toISOString(), experimental: true, server: options.irodori.server };
   } else {
-    const estimatedCostUsd = options.engine === "gemini-tts"
-      ? estimateGeminiTtsCostUsd(readingText.length) : estimateFalCostUsd(readingText);
+    const engine = ttsEngine(options.engine);
+    const estimatedCostUsd = estimateTtsCost(engine, readingText);
     costUsd = estimatedCostUsd;
-    io.logError(`${options.engine} 推定費用: 約 $${estimatedCostUsd}（${readingText.length} 文字）`);
+    io.logError(estimatedCostUsd === null ? `${options.engine} 見積不可（${readingText.length} 文字）` :
+      `${options.engine} 推定費用: 約 $${estimatedCostUsd}（${readingText.length} 文字）`);
     if (!options.yes) {
       if (options.json) printCompactJson({ version: 1, status: "needs_approval", estimate_usd: estimatedCostUsd, chars: readingText.length,
+        ...(estimatedCostUsd === null ? { reason: '見積不可' } : {}),
         ...(options.speed !== null ? { speed_applied: false, warnings } : {}) }, io.log);
       else printJson({ sent: false, engine: options.engine, estimated_cost_usd: estimatedCostUsd,
         reason: "費用承認（--yes）がありません。実リクエストは送信していません。" }, io.log);
       return 2;
     }
-    const falKey = resolveFalKey();
-    // --yes が明示された場合のみ、ここで初めて課金の発生する fal API を呼び出す。
-    if (options.engine === "gemini-tts") {
-      audioBuffer = await synthesizeGeminiTts(readingText, { voice: options.voice, model: "gemini-2.5-flash-tts", style: options.style, falKey });
+    // --yes が明示された場合のみ、ここで初めて課金の発生する API を呼び出す。
+    if (engine.provider !== 'fal') {
+      const name = engine.provider === 'fish-audio' ? 'FISH_AUDIO_API_KEY' : 'GEMINI_API_KEY';
+      const key = readProviderKey(name);
+      if (!key) throw new PublicError(`${name} が未設定です。<AKARI_HOME>/credentials.env に登録してください`, 2);
+      const record = options.profile ? guardedProfile(options.profile, engine) : null;
+      let referenceAudio;
+      if (record && engine.supports.clone === 'per-request') {
+        profileReferenceDataUri(record); // 正本 wav と 20 MB 上限を確認する。
+        if (!record.meta.reference_text) throw new PublicError('声プロファイルの reference_text がありません', 2);
+        referenceAudio = fs.readFileSync(path.join(record.dir, record.meta.reference.file));
+      }
+      const payload = engine.buildPayload({ text: readingText,
+        voice: engine.id === 'gemini-3.8-flash-tts' && record ? record.meta.engines[engine.id].voice_id : options.voice, style: options.style,
+        referenceAudio, referenceText: record?.meta.reference_text });
+      audioBuffer = await synthesizeDirect(engine, payload, key);
+      provenance = { provider: engine.provider, engine: engine.id,
+        model: engine.provider === 'fish-audio' ? 's2.1-pro' : 'gemini-3.8-flash-tts',
+        voice: record ? `profile:${options.profile}` : options.voice,
+        generated_at: new Date().toISOString(), price_verified: engine.price.verified };
+    } else if (options.engine === "gemini-tts") {
+      const falKey = resolveFalKey();
+      audioBuffer = await synthesizeCatalog(engine, engine.buildPayload({ text: readingText, voice: options.voice, style: options.style }), falKey);
       provenance = { provider: "fal", engine: "gemini-2.5-flash-tts", voice: `gemini:${options.voice}`,
-        generated_at: new Date().toISOString(), price_verified: false };
-    } else {
+        generated_at: new Date().toISOString(), price_verified: false, endpoint: engine.endpoint };
+    } else if (options.engine === 'fal-qwen3') {
+      const falKey = resolveFalKey();
       const meta = readProfileMeta(options.profile);
-      audioBuffer = await synthesizeFal(readingText, meta, falKey);
+      audioBuffer = await synthesizeCatalog(engine, engine.buildPayload({ text: readingText, profileMeta: meta }), falKey);
       provenance = { provider: "fal", engine: "qwen-3-tts-1.7b", voice: `profile:${options.profile}`,
-        generated_at: new Date().toISOString() };
+        generated_at: new Date().toISOString(), endpoint: engine.endpoint, price_verified: false };
+    } else {
+      const falKey = resolveFalKey();
+      const record = options.profile && engine.supports.clone !== 'none' ? guardedProfile(options.profile, engine) : null;
+      const audioUrl = record && engine.supports.clone === 'per-request' ? profileReferenceDataUri(record) : undefined;
+      const payload = engine.buildPayload({ text: readingText, voice: options.voice, style: options.style,
+        speed: options.speed, profileMeta: record?.meta, audioUrl });
+      audioBuffer = await synthesizeCatalog(engine, payload, falKey);
+      provenance = { provider: 'fal', engine: engine.id, endpoint: engine.endpoint,
+        voice: record ? `profile:${options.profile}` : options.voice ?? 'default',
+        generated_at: new Date().toISOString(), price_verified: engine.price.verified };
     }
   }
 
@@ -1031,7 +1126,7 @@ async function runGenerate(options, io) {
   if (options.json) printCompactJson({ version: 1, status: "ok", id, path: relativePath,
     duration_s: duration, engine: options.engine, voice: provenance.voice, cost_usd: costUsd,
     applied: options.apply, caption_ref: options.captionRef, provenance, warnings,
-    ...(options.speed !== null ? { speed_applied: ["voicevox", "irodori"].includes(options.engine) } : {}) }, io.log);
+    ...(options.speed !== null ? { speed_applied: ["voicevox", "irodori", 'minimax-2.6-hd'].includes(options.engine) } : {}) }, io.log);
   else printJson(entry, io.log);
   return 0;
 }

package/src/tts-engines.mjs

@@ -0,0 +1,181 @@
+// fal TTS の追加点。価格の verified は fal の課金単位を確認できた場合だけ true にする。
+const geminiVoices = Object.entries({
+  Leda: 'Youthful', Achernar: 'Soft', Achird: 'Friendly', Algenib: 'Gravelly',
+  Algieba: 'Smooth', Alnilam: 'Firm', Aoede: 'Breezy', Autonoe: 'Bright',
+  Callirrhoe: 'Easy-going', Charon: 'Informative', Despina: 'Smooth',
+  Enceladus: 'Breathy', Erinome: 'Clear', Fenrir: 'Excitable', Gacrux: 'Mature',
+  Iapetus: 'Clear', Kore: 'Firm', Laomedeia: 'Upbeat', Orus: 'Firm',
+  Pulcherrima: 'Forward', Puck: 'Upbeat', Rasalgethi: 'Informative',
+  Sadachbia: 'Lively', Sadaltager: 'Knowledgeable', Schedar: 'Even',
+  Sulafat: 'Warm', Umbriel: 'Easy-going', Vindemiatrix: 'Gentle',
+  Zephyr: 'Bright', Zubenelgenubi: 'Casual',
+}).map(([id, description]) => ({ id, label: `${id}（${description}）`, ...(id === 'Leda' ? { default: true } : {}) }));
+
+const price = (unit, value, verified = false, as_of = '2026-09-24') => ({ unit, value, verified, as_of });
+const audio = result => result?.audio?.url;
+const fixture = endpoint => `${endpoint.replaceAll('/', '_')}.json`;
+const row = (id, label, endpoint, details) => ({
+  id, label, provider: 'fal', place: 'cloud', endpoint,
+  openapi_fixture: fixture(endpoint), ...details, parseAudio: audio,
+  request: (payload, key) => ({ url: `https://fal.run/${endpoint}`, options: {
+    method: 'POST', headers: { Authorization: `Key ${key}`, 'Content-Type': 'application/json' }, body: JSON.stringify(payload),
+  } }),
+});
+
+export const FAL_TTS_ENGINES = Object.freeze([
+  row('gemini-tts', 'Gemini 2.5 Flash TTS', 'fal-ai/gemini-tts', {
+    openapi_fixture: 'fal_gemini-tts.json',
+    price: price('usd_per_1000_chars', 0.05, false, '2026-09-22'), voices: geminiVoices,
+    default_voice: 'Leda', supports: { speed: false, style: true, clone: 'none' },
+    buildPayload: ({ text, voice, style }) => ({ prompt: text, voice, model: 'gemini-2.5-flash-tts', output_format: 'mp3',
+      language_code: 'Japanese (Japan)', ...(style ? { style_instructions: style } : {}) }),
+  }),
+  row('fal-qwen3', 'fal Qwen3-TTS', 'fal-ai/qwen-3-tts/text-to-speech/1.7b', {
+    price: price('usd_per_1000_chars', 0.09), voices: null,
+    default_voice: null, supports: { speed: false, style: false, clone: 'registered' },
+    buildPayload: ({ text, profileMeta }) => ({ text, language: 'Japanese',
+      speaker_voice_embedding_file_url: profileMeta.embedding_source_url,
+      reference_text: profileMeta.reference_text, max_new_tokens: 2048 }),
+  }),
+  row('gemini-3.1-flash-tts', 'Gemini 3.1 Flash TTS（既製の声）', 'fal-ai/gemini-3.1-flash-tts', {
+    price: price(null, null), voices: geminiVoices, default_voice: 'Leda',
+    supports: { speed: false, style: true, clone: 'none' },
+    buildPayload: ({ text, voice, style }) => ({ prompt: text, voice, output_format: 'mp3',
+      language_code: 'Japanese (Japan)', ...(style ? { style_instructions: style } : {}) }),
+  }),
+  row('elevenlabs-v3', 'ElevenLabs v3（既製の声）', 'fal-ai/elevenlabs/tts/eleven-v3', {
+    price: price('usd_per_1000_chars', 0.10),
+    voices: ['Rachel', 'Aria', 'Sarah', 'Laura', 'Charlie', 'George', 'River', 'Liam', 'Charlotte', 'Alice']
+      .map(id => ({ id, label: id, ...(id === 'Rachel' ? { default: true } : {}) })),
+    default_voice: 'Rachel', supports: { speed: false, style: false, clone: 'none' },
+    buildPayload: ({ text, voice }) => ({ text, voice, language_code: 'ja' }),
+  }),
+  row('minimax-2.6-hd', 'MiniMax Speech 2.6 HD（自分の声対応）', 'fal-ai/minimax/speech-2.6-hd', {
+    price: price(null, null),
+    voices: ['Wise_Woman', 'Friendly_Person', 'Inspirational_girl', 'Deep_Voice_Man',
+      'Calm_Woman', 'Casual_Guy', 'Lively_Girl', 'Patient_Man', 'Lovely_Girl', 'Elegant_Man']
+      .map(id => ({ id, label: id, ...(id === 'Wise_Woman' ? { default: true } : {}) })),
+    default_voice: 'Wise_Woman', supports: { speed: true, style: false, clone: 'registered' },
+    buildPayload: ({ text, voice, profileMeta, speed }) => ({ prompt: text, language_boost: 'Japanese',
+      output_format: 'url', voice_setting: { voice_id: profileMeta?.engines?.['minimax-2.6-hd']?.custom_voice_id ?? voice,
+        ...(speed != null ? { speed } : {}) } }),
+  }),
+  row('chatterbox', 'Chatterbox 多言語（参照音声対応）', 'fal-ai/chatterbox/text-to-speech/multilingual', {
+    caution: '日本語の読みが不安定です（聞き取りで確かめるのがおすすめ）',
+    price: price('usd_per_1000_chars', 0.025), voices: [{ id: 'japanese', label: '日本語', default: true }], default_voice: 'japanese',
+    supports: { speed: false, style: false, clone: 'per-request' },
+    buildPayload: ({ text, audioUrl }) => ({ text, voice: audioUrl ?? 'japanese',
+      ...(audioUrl ? { custom_audio_language: 'japanese' } : {}) }),
+  }),
+  row('index-tts-2', 'Index TTS 2（参照音声対応）', 'fal-ai/index-tts-2/text-to-speech', {
+    price: price('usd_per_second', 0.002), voices: null, default_voice: null,
+    supports: { speed: false, style: false, clone: 'per-request' },
+    buildPayload: ({ text, audioUrl }) => ({ prompt: text, audio_url: audioUrl }),
+  }),
+]);
+
+export const falTtsEngine = id => FAL_TTS_ENGINES.find(engine => engine.id === id);
+
+const fishVoices = [
+  ['5161d41404314212af1254556477c17d', '元気な女性'],
+  ['46745543e52548238593a3962be77e3a', 'ふうか'],
+  ['63bc41e652214372b15d9416a30a60b4', '元気な女性v2'],
+  ['0089dce5fefb4c6ba9b9f2f0debe1ddc', '落ち着いた女性'],
+  ['45c5d3723c9c42f598e4776dcfd5f02d', '落ち着いた男性'],
+].map(([id, label], index) => ({ id, label, ...(index === 0 ? { default: true } : {}) }));
+
+const fishStyle = style => {
+  const value = style?.trim();
+  if (!value) return '';
+  return /^(?:\[[^\[\]]+\])+$/.test(value) ? value : `[${value}]`;
+};
+
+// Fish の TTSRequest.references はバイナリ音声を含み、公式 OpenAPI は MessagePack 必須とする。
+// 保存済み model ID を作らず正本録音を毎回送るため、声の更新と撤回がローカルの正本に従う。
+function msgpack(value) {
+  const parts = [];
+  const write = (head, data) => { parts.push(Buffer.from(head)); if (data) parts.push(data); };
+  const encode = item => {
+    if (Buffer.isBuffer(item)) { const head = Buffer.alloc(5); head[0] = 0xc6; head.writeUInt32BE(item.length, 1); write(head, item); }
+    else if (typeof item === 'string') { const data = Buffer.from(item); const head = Buffer.alloc(5); head[0] = 0xdb; head.writeUInt32BE(data.length, 1); write(head, data); }
+    else if (Array.isArray(item)) { const head = Buffer.alloc(5); head[0] = 0xdd; head.writeUInt32BE(item.length, 1); write(head); item.forEach(encode); }
+    else if (item && typeof item === 'object') { const entries = Object.entries(item); const head = Buffer.alloc(5); head[0] = 0xdf; head.writeUInt32BE(entries.length, 1); write(head); for (const [key, value] of entries) { encode(key); encode(value); } }
+    else throw new TypeError('MessagePack の値が不正です');
+  };
+  encode(value);
+  return Buffer.concat(parts);
+}
+
+export function pcmToWav(pcm) {
+  if (pcm.length % 2) throw new Error('PCM のバイト数が不正です');
+  const header = Buffer.alloc(44);
+  header.write('RIFF', 0); header.writeUInt32LE(pcm.length + 36, 4); header.write('WAVEfmt ', 8);
+  header.writeUInt32LE(16, 16); header.writeUInt16LE(1, 20); header.writeUInt16LE(1, 22);
+  header.writeUInt32LE(24000, 24); header.writeUInt32LE(48000, 28); header.writeUInt16LE(2, 32);
+  header.writeUInt16LE(16, 34); header.write('data', 36); header.writeUInt32LE(pcm.length, 40);
+  return Buffer.concat([header, pcm]);
+}
+
+export function parseGeminiAudio(result) {
+  const content = result?.steps?.filter(step => step.type === 'model_output').flatMap(step => step.content ?? [])
+    .find(part => part.type === 'audio' && typeof part.data === 'string');
+  if (!content) throw new Error('Gemini API の応答に音声がありません');
+  const bytes = Buffer.from(content.data, 'base64');
+  if (bytes.toString('ascii', 0, 4) === 'RIFF' && bytes.toString('ascii', 8, 12) === 'WAVE') return bytes;
+  if (content.mime_type === 'audio/l16') return pcmToWav(bytes);
+  throw new Error(`Gemini API の音声形式に対応していません: ${content.mime_type ?? 'unknown'}`);
+}
+
+export const DIRECT_TTS_ENGINES = Object.freeze([
+  {
+    id: 'fish-s2.1-pro', label: 'Fish Audio S2.1-Pro', provider: 'fish-audio', place: 'cloud',
+    endpoint: 'https://api.fish.audio/v1/tts', openapi_fixture: 'direct/fish-audio.md',
+    price: price('usd_per_1000_chars', 0.045, true), voices: fishVoices,
+    default_voice: fishVoices[0].id, supports: { speed: false, style: true, clone: 'per-request' },
+    buildPayload: ({ text, voice, style, referenceAudio, referenceText }) => ({
+      text: style?.trim() ? `${fishStyle(style)} ${text}` : text,
+      ...(referenceAudio ? { references: [{ audio: referenceAudio, text: referenceText }] } : { reference_id: voice }),
+      format: 'mp3',
+    }),
+    request: (payload, key) => ({ url: 'https://api.fish.audio/v1/tts', options: {
+      method: 'POST', headers: { Authorization: `Bearer ${key}`, model: 's2.1-pro',
+        'Content-Type': payload.references ? 'application/msgpack' : 'application/json' },
+      body: payload.references ? msgpack(payload) : JSON.stringify(payload),
+    } }),
+  },
+  {
+    id: 'gemini-3.8-flash-tts', label: 'Gemini 3.8 Flash TTS', provider: 'google-ai', place: 'cloud',
+    endpoint: 'https://generativelanguage.googleapis.com/v1beta/interactions', openapi_fixture: 'direct/gemini-tts.md',
+    price: price('usd_per_second', 9 * 25 / 1_000_000, true), voices: geminiVoices,
+    default_voice: 'Leda', supports: { speed: false, style: true, clone: 'registered' },
+    buildPayload: ({ text, voice, style }) => ({ model: 'gemini-3.8-flash-tts', input: [{ type: 'user_input',
+      content: [{ type: 'text', text, ...(style ? { annotations: [{ type: 'speech_metadata', style }] } : {}) }] }],
+      response_format: { type: 'audio', mime_type: 'audio/l16', sample_rate: 24000 },
+      generation_config: { speech_config: [{ voice }] },
+    }),
+    request: (payload, key) => ({ url: 'https://generativelanguage.googleapis.com/v1beta/interactions', options: {
+      method: 'POST', headers: { 'x-goog-api-key': key, 'Content-Type': 'application/json' }, body: JSON.stringify(payload),
+    } }),
+    parseAudio: parseGeminiAudio,
+  },
+]);
+
+export const ttsEngine = id => falTtsEngine(id) ?? DIRECT_TTS_ENGINES.find(engine => engine.id === id);
+
+export const MAX_REFERENCE_BYTES = 20 * 1024 * 1024;
+
+export function referenceDataUri(bytes) {
+  if (!Buffer.isBuffer(bytes)) throw new Error('参照音声のデータが不正です');
+  if (bytes.length > MAX_REFERENCE_BYTES) throw new Error('20 MB 超の参照音声は送信できません');
+  return `data:audio/wav;base64,${bytes.toString('base64')}`;
+}
+
+export function estimateTtsCost(engine, text) {
+  if (engine.id === 'fish-s2.1-pro') return Number((Buffer.byteLength(text, 'utf8') * 15 / 1_000_000).toFixed(6));
+  const { unit, value } = engine.price;
+  if (value == null) return null;
+  if (unit === 'usd_per_1000_chars') return Number((text.length * value / 1000).toFixed(6));
+  if (unit === 'usd_per_second') return Number((text.length / 5 * value).toFixed(6)); // 日本語約 5 字/秒の概算
+  if (unit === 'usd_per_request') return value;
+  return null;
+}