arona-agent 1.0.0

package/src/utils/spawn.ts

@@ -0,0 +1,56 @@
+import { spawn, type SpawnOptions, type ChildProcessWithoutNullStreams } from "child_process";
+
+/**
+ * 跨平台 spawn 封装。
+ *
+ * Node >= 20.12.2 / 18.20.2 / 21.7.3（2024-04-10 起，CVE-2024-27980 安全修复）引入破坏性变更：
+ * Windows 上 spawn() 直接执行 .bat / .cmd 文件、且未设置 `shell: true` 时会直接抛 EINVAL。
+ * 这里检测命令是否为批处理文件，命中时自动补 shell，保证 Windows 端不炸、其他平台零影响。
+ *
+ * 注意：shell 仅在 .bat/.cmd 场景启用，避免对普通 exe 命令引入 shell 解析副作用。
+ * 调用方传入的参数均为内部固定值，无外部不可信输入，注入风险可控。
+ */
+function isBatchFile(command: string): boolean {
+  // 取命令第一个 token（兼容带引号的完整路径）
+  const trimmed = command.trim();
+  let first: string;
+  if (trimmed.startsWith('"')) {
+    const end = trimmed.indexOf('"', 1);
+    first = end === -1 ? trimmed.slice(1) : trimmed.slice(1, end);
+  } else {
+    first = trimmed.split(/\s+/)[0];
+  }
+  return /\.(bat|cmd)$/i.test(first);
+}
+
+export function spawnCompat(
+  command: string,
+  args: string[] = [],
+  options: SpawnOptions = {},
+): ChildProcessWithoutNullStreams {
+  if (process.platform === "win32" && isBatchFile(command)) {
+    return spawn(command, args, { ...options, shell: true }) as ChildProcessWithoutNullStreams;
+  }
+  return spawn(command, args, options) as ChildProcessWithoutNullStreams;
+}
+
+/**
+ * 剔除代理环境变量，返回新对象（不改动原 process.env）。
+ *
+ * Python 侧 websockets/requests 会读 ALL_PROXY/all_proxy 走 SOCKS 代理；用户本机常开 Clash
+ * 为 GitHub 等设 SOCKS 代理，但阿里云百炼 DashScope 是国内服务、应直连。若代理变量泄漏进
+ * Python 子进程，websockets 16.x 会报 "connecting through a SOCKS proxy requires python-socks"。
+ * 故所有 spawn Python 语音脚本（TTS/STT）时都用此函数清洗 env。
+ */
+export function stripProxyEnv(env: NodeJS.ProcessEnv): NodeJS.ProcessEnv {
+  const out: NodeJS.ProcessEnv = { ...env };
+  for (const key of [
+    "ALL_PROXY", "all_proxy",
+    "SOCKS_PROXY", "socks_proxy",
+    "HTTP_PROXY", "http_proxy",
+    "HTTPS_PROXY", "https_proxy",
+  ]) {
+    delete out[key];
+  }
+  return out;
+}

package/src/voice.ts

@@ -0,0 +1,83 @@
+import { runPython } from "./utils/python.ts";
+import { config } from "./config.ts";
+import { getMainAgent, type AgentId } from "./agent_registry.ts";
+import { hasVoice } from "./voices.ts";
+
+let ttsEnabled = config.noVoice ? false : config.ttsEnabled;
+let sttEnabled = config.noVoice ? false : config.sttEnabled;
+
+/** 指定角色是否已有克隆音色（读 voices.json，arona 缺省回退旧 settings.ttsVoice）。 */
+export function hasVoiceFor(agent: AgentId): boolean {
+  return hasVoice(agent);
+}
+
+/** 当前主 Agent 是否已有克隆音色 */
+export function hasCurrentVoice(): boolean {
+  return hasVoiceFor(getMainAgent());
+}
+
+/** 指定角色在当前全局 TTS 开关下是否可朗读（无音色 → 强制静音，不影响 STT）。 */
+export function isTtsEnabledFor(agent: AgentId): boolean {
+  if (config.noVoice) return false;
+  if (!ttsEnabled) return false;
+  if (!hasVoiceFor(agent)) return false;
+  return true;
+}
+
+export function isTtsEnabled(): boolean {
+  return isTtsEnabledFor(getMainAgent());
+}
+
+export function setTtsEnabled(enabled: boolean): void {
+  if (config.noVoice) return;
+  ttsEnabled = enabled;
+}
+
+export function isSttEnabled(): boolean {
+  return sttEnabled;
+}
+
+export function setSttEnabled(enabled: boolean): void {
+  if (config.noVoice) return;
+  sttEnabled = enabled;
+}
+
+/**
+ * Strip markdown formatting and code blocks before sending to TTS.
+ * （供 tts_stream.ts 实时流式管道复用）
+ */
+export function stripMarkdown(text: string): string {
+  return text
+    .replace(/```[\s\S]*?```/g, " (code block) ")
+    .replace(/`[^`]+`/g, "")
+    .replace(/!\[.*?\]\(.*?\)/g, "")
+    .replace(/\[([^\]]+)\]\([^)]+\)/g, "$1")
+    .replace(/#{1,6}\s+/g, "")
+    .replace(/\*{1,2}([^*]+)\*{1,2}/g, "$1")
+    .replace(/\n{3,}/g, "\n\n")
+    .trim();
+}
+
+/**
+ * Listen for speech via Qwen ASR (阿里云百炼), return transcribed text.
+ */
+export async function listen(): Promise<string> {
+  if (!config.sttApiKey) {
+    console.warn("STT: QWEN_STT_API_KEY not configured");
+    return "";
+  }
+
+  try {
+    const text = await runPython("stt.py", [], undefined, {
+      QWEN_WORKSPACE_ID: config.workspaceId,
+      QWEN_STT_API_KEY: config.sttApiKey,
+      QWEN_STT_MODEL: config.sttModel,
+      QWEN_STT_FORMAT: config.sttFormat,
+      QWEN_STT_SAMPLE_RATE: String(config.sttSampleRate),
+    }, 60000);
+    return text;
+  } catch (err) {
+    console.warn(`STT error: ${err instanceof Error ? err.message : err}`);
+    return "";
+  }
+}

package/src/voice_cli.ts

@@ -0,0 +1,198 @@
+// arona voice add [<角色名>] 命令。
+// - 带角色名：读 settings.json 的百炼 Key，克隆该角色音色；已有音色则询问是否重新克隆。
+// - 不带参数：进入与 setup 相同的多选 TUI，仅列未补全音色的角色。
+// 独立进程入口（由 bin/arona.mjs 在 args[0]==="voice" 时 spawn），不依赖 REPL。
+
+import * as readline from "readline";
+import { existsSync, readFileSync } from "fs";
+import chalk from "chalk";
+import { SETTINGS_FILE, verbose } from "./config.ts";
+import { AGENT_IDS, getAgentLabel, type AgentId } from "./agent_registry.ts";
+import { cloneVoice, hasVoice, setVoiceId } from "./voices.ts";
+import { multiSelect } from "./tui_select.ts";
+import { t } from "./locale.ts";
+
+function readSettings(): { ttsApiKey: string; ttsModel: string } {
+  let ttsApiKey = "";
+  let ttsModel = "qwen-audio-3.0-tts-plus";
+  try {
+    if (existsSync(SETTINGS_FILE)) {
+      const s = JSON.parse(readFileSync(SETTINGS_FILE, "utf-8")) as { ttsApiKey?: unknown; ttsModel?: unknown };
+      if (typeof s.ttsApiKey === "string") ttsApiKey = s.ttsApiKey;
+      if (typeof s.ttsModel === "string" && s.ttsModel) ttsModel = s.ttsModel;
+    }
+  } catch {
+    // settings.json 损坏/缺失：走下方缺 Key 报错
+  }
+  return { ttsApiKey, ttsModel };
+}
+
+/** 演示模式：settings.json#demoMode === true 时，音色克隆静默跳过（等 5s 不写 voices.json）。 */
+function readDemoMode(): boolean {
+  try {
+    if (existsSync(SETTINGS_FILE)) {
+      const s = JSON.parse(readFileSync(SETTINGS_FILE, "utf-8")) as { demoMode?: unknown };
+      return s.demoMode === true;
+    }
+  } catch {
+    // 损坏/缺失：按非演示模式处理
+  }
+  return false;
+}
+
+function isValidAgentId(id: string): id is AgentId {
+  return (AGENT_IDS as readonly string[]).includes(id);
+}
+
+function askYesNo(prompt: string): Promise<boolean> {
+  return new Promise((resolve) => {
+    const rl = readline.createInterface({ input: process.stdin, output: process.stdout });
+    // Ctrl+C 视为"不重新克隆"，立即结束询问（否则 readline 默认清行继续等输入，卡在 y/N）
+    rl.on("SIGINT", () => {
+      console.log();
+      rl.close();
+      resolve(false);
+    });
+    rl.question(prompt, (answer) => {
+      rl.close();
+      resolve(/^y/i.test(answer.trim()));
+    });
+  });
+}
+
+async function cloneOne(agent: AgentId, apiKey: string, model: string, simulate: boolean): Promise<boolean> {
+  if (simulate) {
+    // 演示模式：打印"正在克隆"与"克隆成功"，但不调用 voice_clone.py、不写 voices.json、无演示模式提示。
+    console.log(chalk.cyan(t(
+      `正在克隆 ${getAgentLabel(agent)} 的音色（可能需要 1-2 分钟）...`,
+      `Cloning ${getAgentLabel(agent)}'s voice (may take 1-2 minutes)...`,
+    )));
+    await new Promise((r) => setTimeout(r, 5000));
+    console.log(chalk.green(t(
+      `✓ ${getAgentLabel(agent)} 音色克隆成功`,
+      `✓ ${getAgentLabel(agent)} voice cloned`,
+    )));
+    return true;
+  }
+  console.log(chalk.cyan(t(
+    `正在克隆 ${getAgentLabel(agent)} 的音色（可能需要 1-2 分钟）...`,
+    `Cloning ${getAgentLabel(agent)}'s voice (may take 1-2 minutes)...`,
+  )));
+  try {
+    const voiceId = await cloneVoice(agent, apiKey, model);
+    setVoiceId(agent, voiceId);
+    console.log(chalk.green(t(
+      verbose
+        ? `✓ ${getAgentLabel(agent)} 音色克隆成功（voice_id: ${voiceId}）`
+        : `✓ ${getAgentLabel(agent)} 音色克隆成功`,
+      verbose
+        ? `✓ ${getAgentLabel(agent)} voice cloned (voice_id: ${voiceId})`
+        : `✓ ${getAgentLabel(agent)} voice cloned`,
+    )));
+    return true;
+  } catch (err) {
+    const msg = err instanceof Error ? err.message : String(err);
+    console.log(chalk.red(t(
+      `✗ ${getAgentLabel(agent)} 音色克隆失败：${msg}`,
+      `✗ ${getAgentLabel(agent)} voice cloning failed: ${msg}`,
+    )));
+    return false;
+  }
+}
+
+async function run(argv: string[]): Promise<void> {
+  // --verbose 可与角色名同栈传入（`arona voice add --verbose`）：先剔除，避免被当成角色名
+  const args = argv.filter((a) => a !== "--verbose");
+  const sub = args[0];
+  if (sub !== "add") {
+    console.log(chalk.yellow(t(
+      "用法：arona voice add [<角色名>]",
+      "Usage: arona voice add [<character-name>]",
+    )));
+    process.exit(1);
+  }
+
+  const { ttsApiKey, ttsModel } = readSettings();
+  const demoMode = readDemoMode();
+  if (!demoMode && !ttsApiKey) {
+    console.log(chalk.red(t(
+      "未找到百炼 API Key（settings.json 的 ttsApiKey）。请先运行 arona setup 配置。",
+      "Bailian API Key not found (settings.json ttsApiKey). Run `arona setup` first.",
+    )));
+    process.exit(1);
+  }
+
+  const name = args[1];
+
+  if (name) {
+    // 指定单个角色
+    if (!isValidAgentId(name)) {
+      console.log(chalk.red(t(
+        `未知角色 "${name}"。可用角色：${AGENT_IDS.join("、")}`,
+        `Unknown character "${name}". Available: ${AGENT_IDS.join(", ")}`,
+      )));
+      process.exit(1);
+    }
+    // 演示模式不询问是否重新克隆，直接静默跳过。
+    if (!demoMode && hasVoice(name)) {
+      const again = await askYesNo(t(
+        `角色 ${getAgentLabel(name)} 已存在音色，是否重新克隆？(y/N) `,
+        `Character ${getAgentLabel(name)} already has a voice. Re-clone? (y/N) `,
+      ));
+      if (!again) {
+        console.log(chalk.cyan(t("已取消。", "Cancelled.")));
+        return;
+      }
+    }
+    await cloneOne(name, ttsApiKey, ttsModel, demoMode);
+    return;
+  }
+
+  // 无参：TUI 展示全部角色（主 Agent + 子 Agent）。
+  // 演示模式：阿洛娜强制显示为已克隆（锁定、不重克隆）；普拉娜/砂狼白子/小鸟游星野强制显示为未克隆（可选）。
+  const options = AGENT_IDS.map((id) => {
+    if (demoMode && id === "arona") {
+      return { id, label: `${getAgentLabel(id)}${t("（已克隆）", " (cloned)")}`, locked: true };
+    }
+    if (demoMode) {
+      return { id, label: getAgentLabel(id), locked: false };
+    }
+    return {
+      id,
+      label: hasVoice(id) ? `${getAgentLabel(id)}${t("（已克隆）", " (cloned)")}` : getAgentLabel(id),
+      locked: hasVoice(id),
+    };
+  });
+  const initial = new Set<string>();
+  const selected = await multiSelect(
+    t("选择要补全音色的角色", "Select characters to add voices"),
+    options,
+    initial,
+    t(
+      "  ↑/↓ 切换 · 空格选中 [*] · 回车克隆 · Esc 取消",
+      "  ↑/↓ move · Space select [*] · Enter clone · Esc cancel",
+    ),
+  );
+
+  if (selected === null) {
+    console.log(chalk.cyan(t("已取消。", "Cancelled.")));
+    return;
+  }
+  if (selected.size === 0) {
+    console.log(chalk.cyan(t("未选择任何角色。", "No character selected.")));
+    return;
+  }
+
+  for (const id of AGENT_IDS) {
+    if (!selected.has(id)) continue;
+    await cloneOne(id, ttsApiKey, ttsModel, demoMode);
+  }
+}
+
+run(process.argv.slice(2)).catch((err) => {
+  console.error(chalk.red(t(
+    `voice add 错误：${err instanceof Error ? err.message : err}`,
+    `voice add error: ${err instanceof Error ? err.message : err}`,
+  )));
+  process.exit(1);
+});

package/src/voices.ts

@@ -0,0 +1,135 @@
+// 每角色独立音色注册表（声音复刻 voice_id）。
+// 存储到 ~/.arona/voices.json（不塞进 settings.json——多角色 voice_id 放一个字段不现实）。
+// 结构：{ "arona": "<voice_id>", "plana": "<voice_id>" }
+// 运行时 TTS 按当前主 Agent 的 voice_id 判定：无音色 → 强制静音（见 voice.ts）。
+
+import { existsSync, readFileSync, writeFileSync } from "fs";
+import { join } from "path";
+import { ARONA_DIR, SETTINGS_FILE, PROJECT_ROOT } from "./config.ts";
+import { AGENT_IDS, type AgentId } from "./agent_registry.ts";
+import { runPython } from "./utils/python.ts";
+
+export const VOICES_FILE = join(ARONA_DIR, "voices.json");
+
+/** 每个角色的音色源文件路径（声音复刻上传的音频） */
+export const VOICE_AUDIO: Record<AgentId, string> = {
+  arona: join(PROJECT_ROOT, "assets", "blue-archive", "arona", "voice.mp3"),
+  plana: join(PROJECT_ROOT, "assets", "blue-archive", "plana", "voice.mp3"),
+  shiroko: join(PROJECT_ROOT, "assets", "blue-archive", "shiroko", "voice.mp3"),
+  hoshino: join(PROJECT_ROOT, "assets", "blue-archive", "hoshino", "voice.mp3"),
+};
+
+function loadVoices(): Record<string, string> {
+  if (!existsSync(VOICES_FILE)) return {};
+  try {
+    const parsed = JSON.parse(readFileSync(VOICES_FILE, "utf-8"));
+    if (parsed && typeof parsed === "object" && !Array.isArray(parsed)) {
+      return parsed as Record<string, string>;
+    }
+    return {};
+  } catch {
+    return {};
+  }
+}
+
+/**
+ * 读取指定角色的 voice_id。
+ * 优先 voices.json；arona 缺省时回退读旧 settings.json#ttsVoice（向后兼容存量用户）。
+ */
+export function getVoiceId(agent: AgentId): string {
+  const voices = loadVoices();
+  const v = voices[agent];
+  if (typeof v === "string" && v) return v;
+
+  if (agent === "arona") {
+    try {
+      if (existsSync(SETTINGS_FILE)) {
+        const s = JSON.parse(readFileSync(SETTINGS_FILE, "utf-8")) as { ttsVoice?: unknown };
+        if (typeof s.ttsVoice === "string" && s.ttsVoice) return s.ttsVoice;
+      }
+    } catch {
+      // settings.json 损坏/不可读：忽略
+    }
+  }
+  return "";
+}
+
+/** 指定角色是否已克隆音色。 */
+export function hasVoice(agent: AgentId): boolean {
+  return getVoiceId(agent) !== "";
+}
+
+/** 写回指定角色的 voice_id（read-modify-write，保留其它角色）。 */
+export function setVoiceId(agent: AgentId, voiceId: string): void {
+  const voices = loadVoices();
+  voices[agent] = voiceId;
+  writeFileSync(VOICES_FILE, JSON.stringify(voices, null, 2) + "\n");
+}
+
+/** 尚缺音色的全部角色列表（主 Agent + 子 Agent）。 */
+export function getMissingAgents(): AgentId[] {
+  return AGENT_IDS.filter((id) => !hasVoice(id));
+}
+
+/**
+ * 克隆指定角色的音色，返回 voice_id。
+ * 复用 python/voice_clone.py：ARONA_VOICE_AUDIO + ARONA_VOICE_PREFIX 指定音频与角色前缀。
+ */
+export async function cloneVoice(agent: AgentId, apiKey: string, model: string): Promise<string> {
+  const out = await runPython(
+    "voice_clone.py",
+    [],
+    undefined,
+    {
+      QWEN_TTS_API_KEY: apiKey,
+      QWEN_TTS_MODEL: model,
+      ARONA_VOICE_AUDIO: VOICE_AUDIO[agent],
+      ARONA_VOICE_PREFIX: agent,
+    },
+    360000, // 6 min：python 侧轮询等待上限 5min + upload/create 耗时，Node 预算必须大于
+            // python 侧总耗时，否则服务端克隆已成功但本地先判超时（上传文件也不会被清理）
+  );
+
+  let parsed: { voice_id?: string; error?: string };
+  try {
+    parsed = JSON.parse(out);
+  } catch {
+    throw new Error(`invalid voice_clone output: ${out}`);
+  }
+  if (parsed.voice_id) return parsed.voice_id;
+  throw new Error(parsed.error || "voice_clone returned no voice_id");
+}
+
+/**
+ * 旧格式迁移：settings.json 直接存 voice_id（单角色时代）→ voices.json 新格式。
+ * 动作：settings.json#ttsVoice 非空且 voices.json 尚无 arona → 写入 voices.json.arona；
+ * 然后删除 settings.json#ttsVoice 完成清理（避免每次启动重复检测）。
+ * 幂等：无旧字段直接返回。失败静默（getVoiceId 的 legacy 回退仍兜底）。
+ */
+function migrateLegacyVoice(): void {
+  try {
+    if (!existsSync(SETTINGS_FILE)) return;
+    const settings = JSON.parse(readFileSync(SETTINGS_FILE, "utf-8")) as Record<string, unknown>;
+    const legacy = settings.ttsVoice;
+    if (typeof legacy !== "string" || !legacy) return;
+
+    const voices = loadVoices();
+    let changed = false;
+    if (typeof voices.arona !== "string" || !voices.arona) {
+      voices.arona = legacy;
+      changed = true;
+    }
+    if (changed) {
+      writeFileSync(VOICES_FILE, JSON.stringify(voices, null, 2) + "\n");
+    }
+    delete settings.ttsVoice;
+    writeFileSync(SETTINGS_FILE, JSON.stringify(settings, null, 2) + "\n");
+  } catch {
+    // 迁移失败静默：不影响启动，getVoiceId 的 legacy 回退仍能读到旧字段
+  }
+}
+
+// 模块加载时执行一次：任何入口（arona / arona setup / arona voice add / 未来一切命令）
+// 只要加载本模块就自动完成旧格式 → voices.json 迁移。setup / voice add 读 voices.json 之前
+// 迁移已就位。
+migrateLegacyVoice();