@oadank/dsh-input-tools 0.3.3 → 0.3.4

package/lib/index.js

@@ -20,7 +20,7 @@
  */
 
 import { createHash, randomUUID } from 'node:crypto'
-import { mkdir, open, readFile, unlink, writeFile } from 'node:fs/promises'
+import { mkdir, open, readFile, unlink, writeFile, copyFile } from 'node:fs/promises'
 import { constants, readFileSync } from 'node:fs'
 import { homedir } from 'node:os'
 import { fileURLToPath } from 'node:url'
@@ -34,6 +34,44 @@ const inject = ['tools', 'webServer']
 
 export { name, inject }
 
+// ──────────────────────────────────────────────────────────────
+// [0.3.4] 自带素材（下载即用）：克隆样本 + VoiceDesign 示例音频打进 npm 包 assets/，
+// 首次加载自动拷贝到 DSH_HOME 并注册，不再依赖"手动上传/在线生成"。
+// ──────────────────────────────────────────────────────────────
+const PLUGIN_ROOT = join(fileURLToPath(import.meta.url), '..', '..') // .../dsh-input-tools
+const ASSETS_DIR = join(PLUGIN_ROOT, 'assets')
+const BUNDLED_CLONE_ID = '8da38fcc-b041-4f5b-86b9-901956016f89'
+const BUNDLED_CLONE_SAMPLE = {
+  id: BUNDLED_CLONE_ID,
+  name: '小团团(60秒长样本)',
+  context: '一个魔性的少女萝莉音，说话自带沙雕搞怪和无厘头气质，像在撒娇又像在耍宝，情绪起伏很大：前一句还奶声奶气地撒娇卖萌，后一句就突然拔高音量夸张卖惨耍赖，再下一秒又贱兮兮地坏笑。尾音拖长上扬，带着气音和魔性笑声，喜欢用「臭猪」「你凶我」「哼」「嘿嘿嘿」这类咋咋呼呼的用词，语速忽快忽慢、节奏跳跃，吐字软糯清晰，傻白甜又可爱，让人听了忍不住想笑',
+}
+const VOICE_DESIGN_SAMPLE_KEYS = ['asmr', 'docu', 'elder']
+
+let bundledInitDone = false
+/** 首次加载把自带素材落地到 DSH_HOME：克隆样本 mp3 拷贝 + 首次安装自动注册小团团。 */
+async function ensureBundledAssets(config, parsed) {
+  if (bundledInitDone) return config
+  bundledInitDone = true
+  try {
+    const homeDir = process.env.DSH_HOME ?? join(homedir(), '.dsh')
+    const cloneDir = join(homeDir, 'voiceclone-samples')
+    const dstClone = join(cloneDir, BUNDLED_CLONE_ID + '.mp3')
+    try {
+      await mkdir(cloneDir, { recursive: true })
+      await copyFile(join(ASSETS_DIR, 'voiceclone-samples', BUNDLED_CLONE_ID + '.mp3'), dstClone)
+    } catch { /* 包内素材缺失或拷贝失败：跳过（不阻塞启动） */ }
+    // 仅"首次安装"（配置里还没有 voiceclone 键）时注册自带样本；用户删光的 [] 不强制
+    const parsedHasClone = parsed !== null && typeof parsed === 'object' && parsed.engines?.voiceclone !== undefined
+    const samples = config?.engines?.voiceclone?.samples
+    if (!parsedHasClone && (!Array.isArray(samples) || samples.length === 0)) {
+      config.engines.voiceclone = { ...config.engines.voiceclone, enabled: true, samples: [{ ...BUNDLED_CLONE_SAMPLE, path: dstClone }] }
+      await saveVoiceConfig(config)
+    }
+  } catch { /* 初始化失败不阻塞 */ }
+  return config
+}
+
 // ──────────────────────────────────────────────────────────────
 // 配置中心：~/.dsh/voice-config.json
 // ──────────────────────────────────────────────────────────────
@@ -95,6 +133,8 @@ async function loadVoiceConfig() {
     parsed = JSON.parse(await readFile(CONFIG_PATH, 'utf8'))
   } catch { /* 首次无配置 */ }
   cachedConfig = deepMerge(defaultVoiceConfig(), parsed)
+  // [0.3.4] 自带素材初始化（拷贝克隆样本 + 首次安装自动注册小团团）
+  await ensureBundledAssets(cachedConfig, parsed)
   // 环境变量覆盖（兼容旧配置；显式配置值优先于 env）
   const env = process.env
   if (env.TTS_XIAOMI_KEY !== undefined && cachedConfig.engines.xiaomi.apiKey === '') cachedConfig.engines.xiaomi.apiKey = env.TTS_XIAOMI_KEY
@@ -961,8 +1001,8 @@ async function apply(ctx) {
                 return sendJson(res, 404, { ok: false, error: '安装脚本不存在' })
               }
             }
-            // VoiceDesign 官方示例音频：用真实的 MiMo-V2.5-TTS-VoiceDesign 模型生成（instruct=音色描述 / text=正文），缓存到 ~/.dsh/voice-design-samples/
-            // 注意：必须走 voicedesign 模型（user=音色描述，无 voice 参数）才是真音色，绝不能降级到 edge 假造
+            // VoiceDesign 官方示例音频：[0.3.4] 优先读插件包自带素材（assets/，mp3 下载即用），
+            // 包内缺失才回退到"小米模型在线生成 + 缓存到 ~/.dsh/voice-design-samples/ 的 wav"。
             if (url.pathname === '/asr/voice-design-samples' && req.method === 'GET') {
               const cfg = await loadVoiceConfig()
               const homeDir = process.env.DSH_HOME ?? join(homedir(), '.dsh')
@@ -986,20 +1026,29 @@ async function apply(ctx) {
               ]
               const results = []
               for (const s of samples) {
-                const wavPath = join(sampleDir, `${s.key}.wav`)
-                let wavBytes = null
-                try { wavBytes = await readFile(wavPath) } catch { /* 首次/缓存失效 → 用真实模型重新生成 */ }
-                if (wavBytes === null) {
+                // 1) 包内自带 mp3（首选，下载即用）
+                let bytes = null
+                try { bytes = await readFile(join(ASSETS_DIR, 'voice-design-samples', `${s.key}.mp3`)) } catch { /* 包内无 → 回退 */ }
+                let mediaType = 'audio/mpeg'
+                if (bytes === null) {
+                  // 2) 缓存 wav（在线生成过）
+                  const wavPath = join(sampleDir, `${s.key}.wav`)
+                  try { bytes = await readFile(wavPath) } catch { /* 继续回退 */ }
+                  mediaType = 'audio/wav'
+                }
+                if (bytes === null) {
+                  // 3) 在线生成（key 缺失会失败，示例跳过）
                   try {
                     const syn = await synthesizeXiaomiVoiceDesign(s.text, { context: s.instruct }, cfg)
                     if (syn === null) throw new Error('voicedesign synth failed')
-                    wavBytes = Buffer.from(syn.data)
+                    bytes = Buffer.from(syn.data)
+                    mediaType = 'audio/wav'
                     await mkdir(sampleDir, { recursive: true })
-                    await writeFile(wavPath, wavBytes)
-                  } catch { /* 模型生成失败（如 key 缺失）→ 该示例跳过，绝不降级假音频 */ }
+                    await writeFile(join(sampleDir, `${s.key}.wav`), bytes)
+                  } catch { /* 模型生成失败 → 跳过 */ }
                 }
-                if (wavBytes !== null) {
-                  results.push({ key: s.key, title: s.title, mediaType: 'audio/wav', data: wavBytes.toString('base64') })
+                if (bytes !== null) {
+                  results.push({ key: s.key, title: s.title, mediaType, data: bytes.toString('base64') })
                 }
               }
               return sendJson(res, 200, { ok: results.length > 0, samples: results })

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@oadank/dsh-input-tools",
-  "version": "0.3.3",
+  "version": "0.3.4",
   "private": false,
   "type": "module",
   "main": "lib/index.js",
@@ -13,6 +13,7 @@
   "files": [
     "lib",
     "scripts",
+    "assets",
     "README.md",
     "cordis.patch.yml"
   ],

package/scripts/install-asr.ps1

@@ -168,7 +168,7 @@ const server = http.createServer(async (req, res) => {
     res.writeHead(404); res.end();
   }
 });
-server.listen(PORT, () => console.log('[ASR] listening on ' + PORT));
+server.listen(PORT, '127.0.0.1', () => console.log('[ASR] listening on 127.0.0.1:' + PORT));
 "@
 # 服务脚本用 .cjs：插件包 package.json 声明了 "type":"module"，.js 会被当 ESM 解析导致 require 报错
 # （2026-08-21 XDN 实测：asr-service.js 抛 ReferenceError: require is not defined）