dsh-liuying-voice 1.1.0 → 1.2.0

package/lib/config.js

@@ -1,4 +1,4 @@
-/** 设置存储：$DSH_HOME/liuying-voice/settings.json（无敏感值）。 */
+/** 设置存储：$DSH_HOME/liuying-voice/settings.json（无敏感值）。 */
 import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
 import os from "node:os";
 import path from "node:path";
@@ -59,6 +59,11 @@ export function describeSettings(settings) {
     pitch: Number.isFinite(Number(settings.pitch)) ? Math.min(12, Math.max(-12, Number(settings.pitch))) : 0,
     protect: Number.isFinite(Number(settings.protect)) ? Math.min(0.5, Math.max(0, Number(settings.protect))) : 0.33,
     outDir: pick(settings.outDir),
+    engine: settings.engine === "gptsovits" ? "gptsovits" : "rvc",
+    gptApi: pick(settings.gptApi),
+    gptRef: pick(settings.gptRef),
+    gptRefText: pick(settings.gptRefText),
+    gptSpeed: Number.isFinite(Number(settings.gptSpeed)) ? Math.min(2, Math.max(0.5, Number(settings.gptSpeed))) : 1.0,
     qqGuard: settings.qqGuard !== false,
     snowLumaDir: pick(settings.snowLumaDir),
     snowLumaNode: pick(settings.snowLumaNode),

package/lib/index.js

@@ -363,6 +363,27 @@ function mountSettings(ctx) {
   });
 }
 
+function mountSettingsPage(ctx) {
+  const web = ctx.get("webServer");
+  if (!web?.register) return null;
+  return web.register({
+    kind: "exact",
+    path: "/liuying-voice",
+    handler: async (req, res) => {
+      noteDshPort(req.headers?.host);
+      try {
+        const html = readFileSync(path.join(PLUGIN_ROOT, "lib", "settings-page.html"), "utf8");
+        res.statusCode = 200;
+        res.setHeader("content-type", "text/html; charset=utf-8");
+        res.end(html);
+      } catch {
+        res.statusCode = 500;
+        res.end("settings-page.html 缺失");
+      }
+    },
+  });
+}
+
 export async function apply(ctx, config = {}) {
   diag(`apply 开始；webServer=${ctx.get("webServer") ? "有" : "无"}；tools=${ctx.get("tools") ? "有" : "无"}`);
   // 一次性把可能影响 python 子进程的环境变量记下来（排查 edge-tts 用）
@@ -391,6 +412,19 @@ export async function apply(ctx, config = {}) {
   } catch (error) {
     log(`设置接口挂载失败：${error?.message ?? error}`);
   }
+  try {
+    const dispose = mountSettingsPage(ctx);
+    if (typeof dispose === "function") {
+      const prev = disposeSettings;
+      disposeSettings = () => {
+        if (typeof prev === "function") prev();
+        dispose();
+      };
+    }
+    log("设置页面已挂载：/liuying-voice");
+  } catch (error) {
+    log(`设置页面挂载失败：${error?.message ?? error}`);
+  }
   try {
     const dispose = mountMemory(ctx);
     disposeMemory = typeof dispose === "function" ? dispose : null;

package/lib/settings-page.html

@@ -0,0 +1,126 @@
+<!doctype html>
+<html lang="zh-CN">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width,initial-scale=1">
+<title>流萤语音 · 设置</title>
+<style>
+  :root { color-scheme: dark; }
+  * { box-sizing: border-box; }
+  body { margin: 0; font: 14px/1.6 "Segoe UI", "Microsoft YaHei", sans-serif; background: #0f1417; color: #d7e4e6; }
+  .wrap { max-width: 760px; margin: 0 auto; padding: 28px 20px 60px; }
+  h1 { font-size: 20px; margin: 0 0 4px; letter-spacing: .04em; }
+  .sub { color: #6f8a8e; font-size: 12px; margin-bottom: 24px; }
+  .card { background: #161d21; border: 1px solid #243036; border-radius: 10px; padding: 18px 20px; margin-bottom: 16px; }
+  .card h2 { font-size: 13px; color: #8fd8c8; margin: 0 0 14px; font-weight: 600; letter-spacing: .06em; }
+  .row { display: flex; align-items: center; gap: 12px; margin: 10px 0; }
+  .row label { flex: 0 0 150px; color: #9db5b9; font-size: 13px; }
+  .row input[type=text], .row input[type=number], .row select, .row textarea {
+    flex: 1; background: #0f1417; color: #d7e4e6; border: 1px solid #2a383f; border-radius: 6px;
+    padding: 7px 10px; font: inherit; outline: none;
+  }
+  .row input:focus, .row select:focus, .row textarea:focus { border-color: #4db6a0; }
+  .row input[type=checkbox] { width: 18px; height: 18px; accent-color: #4db6a0; }
+  .hint { color: #5f7478; font-size: 12px; margin: -4px 0 10px 162px; }
+  .btns { display: flex; gap: 10px; margin-top: 22px; }
+  button { font: inherit; padding: 9px 20px; border-radius: 7px; border: 1px solid #2a383f; background: #1c262b; color: #cfe3e6; cursor: pointer; }
+  button.primary { background: #1f6f5f; border-color: #2b8a76; color: #eafaf5; }
+  button:hover { filter: brightness(1.15); }
+  .status { margin-top: 14px; font-size: 13px; min-height: 20px; color: #8fd8c8; }
+  .status.err { color: #e08a8a; }
+</style>
+</head>
+<body>
+<div class="wrap">
+  <h1>流萤语音</h1>
+  <div class="sub">dsh-liuying-voice · 独立设置页</div>
+
+  <form id="form">
+    <div class="card">
+      <h2>通用</h2>
+      <div class="row"><label>启用</label><input type="checkbox" id="enabled"></div>
+      <div class="row"><label>发送模式</label>
+        <select id="sendMode"><option value="mix">混合（文字+语音）</option><option value="voice">只用语音</option><option value="text">只用文字</option></select>
+      </div>
+      <div class="row"><label>语音比例（mix）</label><input type="number" id="voiceRatio" min="0" max="1" step="0.1"></div>
+    </div>
+
+    <div class="card">
+      <h2>引擎</h2>
+      <div class="row"><label>语音引擎</label>
+        <select id="engine">
+          <option value="rvc">RVC（本地 edge-tts + 变声，快）</option>
+          <option value="gptsovits">GPT-SoVITS（更自然，需 9880 服务）</option>
+        </select>
+      </div>
+      <div class="row"><label>GPT-SoVITS 地址</label><input type="text" id="gptApi" placeholder="http://127.0.0.1:9880/"></div>
+      <div class="row"><label>参考音频</label><input type="text" id="gptRef" placeholder="留空用插件自带 assets/firefly-ref-hf.wav"></div>
+      <div class="row"><label>参考文本</label><textarea id="gptRefText" rows="2"></textarea></div>
+      <div class="row"><label>语速</label><input type="number" id="gptSpeed" min="0.5" max="2" step="0.05"></div>
+      <div class="hint">GPT-SoVITS 首次使用会自动下载模型，需要本地 9880 服务在跑。</div>
+    </div>
+
+    <div class="card">
+      <h2>RVC（旧引擎）</h2>
+      <div class="row"><label>Python</label><input type="text" id="python"></div>
+      <div class="row"><label>Applio 目录</label><input type="text" id="applio"></div>
+      <div class="row"><label>模型 .pth</label><input type="text" id="model"></div>
+      <div class="row"><label>基础音色</label><input type="text" id="ttsVoice"></div>
+      <div class="row"><label>语速（负=慢）</label><input type="number" id="ttsRate" step="1"></div>
+      <div class="row"><label>音高</label><input type="number" id="pitch" step="1"></div>
+      <div class="row"><label>保护</label><input type="number" id="protect" min="0" max="0.5" step="0.01"></div>
+    </div>
+
+    <div class="btns">
+      <button type="submit" class="primary">保存</button>
+      <button type="button" id="reload">重新载入</button>
+    </div>
+    <div class="status" id="status"></div>
+  </form>
+</div>
+<script>
+const API = "/liuying-voice/settings";
+const FIELDS = ["enabled","sendMode","voiceRatio","engine","gptApi","gptRef","gptRefText","gptSpeed",
+  "python","applio","model","ttsVoice","ttsRate","pitch","protect"];
+const $ = (id) => document.getElementById(id);
+function setStatus(msg, err) { const s = $("status"); s.textContent = msg; s.className = "status" + (err ? " err" : ""); }
+
+async function load() {
+  setStatus("读取中…");
+  try {
+    const r = await fetch(API, { cache: "no-store" });
+    const s = await r.json();
+    for (const k of FIELDS) {
+      const el = $(k);
+      if (!el) continue;
+      if (el.type === "checkbox") el.checked = s[k] !== false;
+      else el.value = s[k] ?? "";
+    }
+    setStatus("已载入");
+  } catch (e) { setStatus("读取失败：" + e.message, true); }
+}
+
+$("form").addEventListener("submit", async (ev) => {
+  ev.preventDefault();
+  const body = {};
+  for (const k of FIELDS) {
+    const el = $(k);
+    if (!el) continue;
+    if (el.type === "checkbox") body[k] = el.checked;
+    else if (el.type === "number") body[k] = Number(el.value);
+    else body[k] = el.value;
+  }
+  setStatus("保存中…");
+  try {
+    const r = await fetch(API, { method: "POST", headers: { "content-type": "application/json" }, body: JSON.stringify(body) });
+    const j = await r.json();
+    if (!j.ok) throw new Error(j.error || "保存失败");
+    setStatus("已保存");
+  } catch (e) { setStatus("保存失败：" + e.message, true); }
+});
+
+$("reload").addEventListener("click", load);
+load();
+</script>
+</body>
+</html>

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "dsh-liuying-voice",
-  "version": "1.1.0",
-  "description": "流萤语音插件 for DeepSeek Harness：本地 edge-tts + RVC / GPT-SoVITS 双引擎，自带流萤模型与参考音频，让 agent 用「流萤」的声音说话。开源 MIT。",
+  "version": "1.2.0",
+  "description": "流萤语音插件 for DeepSeek Harness：本地 edge-tts + RVC / GPT-SoVITS 双引擎，自带流萤参考音频、一键环境安装脚本和独立设置页面。开源 MIT。",
   "keywords": [
     "deepseek-harness",
     "dsh",
@@ -29,6 +29,7 @@
     "assets",
     "scripts/*.py",
     "scripts/*.mjs",
+    "scripts/*.ps1",
     "memory",
     "test",
     "cordis.patch.yml",

package/scripts/install-gpt-sovits.ps1

@@ -0,0 +1,115 @@
+# ============================================================
+# GPT-SoVITS 简易环境安装脚本（CPU 推理，配合 dsh-liuying-voice 使用）
+#
+# 做什么：
+#   1. 检查 Python 3.10/3.11
+#   2. 下载 GPT-SoVITS 源码
+#   3. 创建虚拟环境并安装推理依赖
+#   4. 下载 hubert / bert 预训练
+#   5. 下载流萤 GPT/SoVITS 权重
+#   6. 写入 CPU 推理配置并生成启动脚本
+#
+# 用法（PowerShell）：
+#   powershell -ExecutionPolicy Bypass -File install-gpt-sovits.ps1
+# ============================================================
+$ErrorActionPreference = 'Stop'
+$repoUrl = 'https://github.com/RVC-Boss/GPT-SoVITS'
+$pipIndex = 'https://mirrors.aliyun.com/pypi/simple'
+$env:HF_ENDPOINT = 'https://hf-mirror.com'
+$env:PYTHONIOENCODING = 'utf-8'
+$root = Join-Path $env:USERPROFILE '.dsh\gpt-sovits'
+
+function Step($m) { Write-Host "`n== $m" -ForegroundColor Cyan }
+
+Step "安装目录：$root"
+New-Item -ItemType Directory -Path $root -Force | Out-Null
+
+# 1. Python
+$python = $null
+foreach ($cand in @('python3.11', 'python3.10', 'python')) {
+    $cmd = Get-Command $cand -ErrorAction SilentlyContinue
+    if ($cmd) { $python = $cmd.Source; break }
+}
+if (-not $python) { throw "需要 Python 3.10/3.11（GPT-SoVITS 依赖），请先安装并加入 PATH" }
+Write-Host "Python: $python"
+
+# 2. 源码
+$repo = Join-Path $root 'GPT-SoVITS'
+if (-not (Test-Path (Join-Path $repo 'api_v2.py'))) {
+    Step '下载 GPT-SoVITS 源码（约 60 MB，慢时请耐心等）'
+    git clone --depth 1 $repoUrl $repo
+}
+if (-not (Test-Path (Join-Path $repo 'api_v2.py'))) { throw "源码下载失败，请检查网络或代理" }
+
+# 3. 虚拟环境
+$venv = Join-Path $root 'venv'
+if (-not (Test-Path (Join-Path $venv 'Scripts\python.exe'))) {
+    Step '创建虚拟环境'
+    & $python -m venv $venv
+}
+$py = Join-Path $venv 'Scripts\python.exe'
+& $py -m pip install -U pip -i $pipIndex
+
+Step '安装推理依赖（阿里镜像）'
+& $py -m pip install -r (Join-Path $repo 'requirements.txt') -i $pipIndex
+& $py -m pip install torch torchaudio --index-url https://download.pytorch.org/whl/cpu
+
+# 4. 预训练
+$pretrained = Join-Path $repo 'GPT_SoVITS\pretrained_models'
+Step '下载 hubert / bert 预训练（约 1.7 GB，走 HF 镜像）'
+& $py -c "from huggingface_hub import snapshot_download; snapshot_download('hfl/chinese-hubert-base', local_dir=r'$pretrained\chinese-hubert-base')"
+& $py -c "from huggingface_hub import snapshot_download; snapshot_download('hfl/chinese-roberta-wwm-ext-large', local_dir=r'$pretrained\chinese-roberta-wwm-ext-large')"
+
+# 5. 流萤模型
+$gw = Join-Path $repo 'GPT_weights'
+$sw = Join-Path $repo 'SoVITS_weights'
+New-Item -ItemType Directory -Path $gw, $sw -Force | Out-Null
+Step '下载流萤 GPT 权重（约 148 MB）'
+$gptFile = Join-Path $gw '流萤-e10.ckpt'
+if (-not (Test-Path $gptFile)) {
+    Invoke-WebRequest -Uri 'https://hf-mirror.com/Shiina-Mahiru/Firefly-Neko/resolve/main/GPT_weights_v2%5C%E6%B5%81%E8%90%A4-e10.ckpt' -OutFile $gptFile -TimeoutSec 3600
+}
+Step '下载流萤 SoVITS 权重（约 81 MB）'
+$sovitsFile = Join-Path $sw '流萤_e15_s810.pth'
+if (-not (Test-Path $sovitsFile)) {
+    Invoke-WebRequest -Uri 'https://hf-mirror.com/Shiina-Mahiru/Firefly-Neko/resolve/main/SoVITS_weights_v2%5C%E6%B5%81%E8%90%A4_e15_s810.pth' -OutFile $sovitsFile -TimeoutSec 3600
+}
+
+# 6. 配置（CPU）
+Step '写入推理配置（CPU 模式）'
+$cfg = Join-Path $repo 'GPT_SoVITS\configs\tts_infer.yaml'
+$cfgText = @"
+custom:
+  bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
+  cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
+  device: cpu
+  is_half: false
+  t2s_weights_path: GPT_weights/流萤-e10.ckpt
+  vits_weights_path: SoVITS_weights/流萤_e15_s810.pth
+default:
+  bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
+  cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
+  device: cpu
+  is_half: false
+  t2s_weights_path: GPT_weights/流萤-e10.ckpt
+  vits_weights_path: SoVITS_weights/流萤_e15_s810.pth
+"@
+[System.IO.File]::WriteAllText($cfg, $cfgText, (New-Object System.Text.UTF8Encoding($false)))
+
+# 7. 启动脚本
+$start = Join-Path $root 'start-api.cmd'
+$startText = @"
+@echo off
+cd /d "$repo"
+set PATH=%PATH%;.\ffmpeg\bin
+set CUDA_VISIBLE_DEVICES=-1
+set OMP_NUM_THREADS=12
+set MKL_NUM_THREADS=12
+"$py" api_v2.py
+"@
+[System.IO.File]::WriteAllText($start, $startText, (New-Object System.Text.UTF8Encoding($false)))
+
+Write-Host "`n完成！" -ForegroundColor Green
+Write-Host "启动 GPT-SoVITS 服务：$start"
+Write-Host "服务地址：http://127.0.0.1:9880/"
+Write-Host "然后在流萤语音插件设置里把 engine 设为 gptsovits。"