dsh-voice-mode 0.5.0 → 0.6.0

package/README.en.md

@@ -11,11 +11,11 @@
 >
 > 中文说明见 [README.md](./README.md)。
 
-> **Version note (0.5.0)**: Edge cloud TTS by default (fast & natural); local TTS (VITS / Kokoro int8, privacy-first) optional + HTTP hardening + model SHA256 pinning form the merged core; `wakeWord` (wake word) and `toolBeep` (tool-call beep) are fully wired; the early fork's `asrModel` (bilingual paraformer) and `punctuate` (neural punctuation) were removed — SenseVoice finalization already adds punctuation, and streaming ASR is fixed to zipformer2. Silence split defaults to 700 ms.
+> **Version note (0.6.0)**: Edge cloud TTS by default (fast & natural); local TTS (VITS / Kokoro, privacy-first) optional + HTTP hardening + model SHA256 pinning form the merged core; Kokoro adds a model-precision choice (`int8` default 109 MB / `fp32` better quality 311 MB); `wakeWord` (wake word) and `toolBeep` (tool-call beep) are fully wired; the early fork's `asrModel` (bilingual paraformer) and `punctuate` (neural punctuation) were removed — SenseVoice finalization already adds punctuation, and streaming ASR is fixed to zipformer2. Silence split defaults to 700 ms.
 
 ## Fork enhancements (this repo)
 
-- **Edge cloud TTS by default; local TTS optional (privacy-first)**: local VITS (Chinese) and local Kokoro int8 (zh+en, native `sherpa-onnx-node` addon, no WASM memory limits) run in an isolated child process;
+- **Edge cloud TTS by default; local TTS optional (privacy-first)**: local VITS (Chinese) and local Kokoro (zh+en, 103 voices; `int8` default ~109 MB / `fp32` ~311 MB for better quality; native `sherpa-onnx-node` addon, no WASM memory limits) run in an isolated child process;
 - **103 Kokoro voices** (F0-measured gender labels, 4 favourite male voices pinned), browsed with a ◀▶ stepper;
 - **Delta transport**: partials upload only the new 0.9 s — long push-to-talk segments finalize in seconds;
 - **Interaction**: a mode-switch button next to the mic (continuous ⇄ hold, persisted); hold mode records only while held;
@@ -34,7 +34,7 @@
   - `toggle` (default) continuous listening: RMS VAD segmentation → streaming zipformer2 ASR (words appear as you speak, live caption preview) → automatic sentence split and send after 700 ms of silence; hold `Ctrl` to force an immediate send
   - `hold` push-to-talk: short tap to enter/exit, **hold the mic button to talk, release to send** (swipe up to cancel, `Esc`/blur abandons the segment; pauses do not split while held, up to 10 min); hold `Ctrl` to record-by-keyboard, release to send
 - **Wake word (optional, off by default)**: after setting `wakeWord`, entering voice mode starts in standby, and recognition only begins once the wake word is spoken (e.g. `你好小D`), preventing accidental triggers
-- **Output pipeline**: only the final answer's `text-delta` is read (reasoning/tool calls are skipped), streamed sentence-by-sentence (Edge cloud by default; local VITS / Kokoro int8 optional) with a live caption overlay at the bottom-right; tool calls trigger a beep; the full text is still written to the chat; in voice mode a spoken-format system prompt is injected (short natural sentences, no Markdown decoration), and the reader side strips markers as well for a smoother listening experience
+- **Output pipeline**: only the final answer's `text-delta` is read (reasoning/tool calls are skipped), streamed sentence-by-sentence (Edge cloud by default; local VITS / Kokoro, int8/fp32, optional) with a live caption overlay at the bottom-right; tool calls trigger a beep; the full text is still written to the chat; in voice mode a spoken-format system prompt is injected (short natural sentences, no Markdown decoration), and the reader side strips markers as well for a smoother listening experience
 - **Barge-in**: three sensitivity levels of voice-onset detection → local mute + host synth queue invalidation (epoch) + running turn cancellation (the half-finished part is kept and naturally flows into your new message)
 - **Lazy model download with progress**: the zipformer2 Chinese streaming model (~160 MB, `.part` resumable) is downloaded on first use with live progress in the status bar; `npm run prefetch` can pre-download it
 - **Resilience**: mic-denied red hint, visible model-download failure, TTS unreachable status hint (auto retry), failed submit keeps the text in the draft, SSE auto-reconnect
@@ -100,7 +100,8 @@ If a wake word is configured, you land in standby first (the status bar prompts
 
 | Key | Default | Description |
 | --- | --- | --- |
-| `ttsEngine` | `edge` | Read-aloud engine: `edge` Microsoft cloud (default, fast) / `vits` local Chinese / `kokoro` local zh+en (int8); **applies live** |
+| `ttsEngine` | `edge` | Read-aloud engine: `edge` Microsoft cloud (default, fast) / `vits` local Chinese / `kokoro` local zh+en; **applies live** |
+| `kokoroModel` | `int8` | Kokoro precision: `int8` (default, 109 MB, CPU/low-bandwidth) / `fp32` (311 MB, better quality, GPU/large memory); same 103 voices; **applies live** |
 | `voice` | per engine | Voice: 5 VITS speakers; 103 Kokoro voices (◀▶ stepper; 62/68/75/76 favourite males pinned); Edge ShortNames below. The inline "试听" button previews it at the current rate |
 | `rate` | `1.0` | Reading speed multiplier (0.5 slow ～ 2.0 fast), **applies live** |
 | `interruptLevel` | `0` | Barge-in sensitivity (host-side VAD frame detection + echo gate): 0 high threshold / 1 medium / 2 low |
@@ -111,7 +112,7 @@ If a wake word is configured, you land in standby first (the status bar prompts
 | `mode` | `toggle` | Interaction mode: `toggle` continuous listening + 700 ms silence split; `hold` push-to-talk, release to send (short tap exits) |
 | `wakeWord` | empty (off) | Wake word (e.g. `你好小D`): speak it after entering to activate, avoiding accidental triggers; empty = off |
 
-Effect timing: `voice`/`rate` take effect **immediately** (TTS hot-swap); the rest apply on the next voice-mode entry. Defaults come from the plugin config (`base` layer) — they follow the config unless explicitly changed.
+Effect timing: `voice`/`rate`/`ttsEngine`/`kokoroModel`/`spokenFormat` take effect **immediately** (TTS hot-swap); the rest apply on the next voice-mode entry. Defaults come from the plugin config (`base` layer) — they follow the config unless explicitly changed.
 
 ### Common voices (full list: `node scripts/list-voices.mjs`)
 

package/README.md

@@ -13,7 +13,7 @@ DeepSeek Harness 语音双工对话模式：会话内一键进入 → 边说边
 
 ![语音模式：实时字幕与状态条](https://raw.githubusercontent.com/qishuilalala/dsh-voice-mode/HEAD/assets/screenshot-voice.png)
 
-> **版本说明（0.4.0）**：朗读默认 **Edge 云端**（快速自然），本地 TTS（VITS/Kokoro）可选（隐私优先）+ HTTP 安全加固 + 模型 SHA256 固定为合入核心；`wakeWord`（唤醒词）与 `toolBeep`（工具提示音）已完整接入；早期 fork 的 `asrModel`（双语 paraformer）与 `punctuate`（神经标点）已移除——SenseVoice 定稿本身已带标点，流式识别固定为 zipformer2。静音断句默认 700 毫秒。
+> **版本说明（0.6.0）**：朗读默认 **Edge 云端**（快速自然），本地 TTS（VITS / Kokoro）可选（隐私优先）+ HTTP 安全加固 + 模型 SHA256 固定为合入核心；Kokoro 新增**模型精度可选**（`int8` 默认 109MB / `fp32` 音质更好 311MB）；`wakeWord`（唤醒词）与 `toolBeep`（工具提示音）已完整接入；早期 fork 的 `asrModel`（双语 paraformer）与 `punctuate`（神经标点）已移除——SenseVoice 定稿本身已带标点，流式识别固定为 zipformer2。静音断句默认 700 毫秒。
 
 ## Fork 增强（本仓库新增）
 
@@ -21,7 +21,7 @@ DeepSeek Harness 语音双工对话模式：会话内一键进入 → 边说边
 
 - **朗读默认 Edge 云端；本地 TTS 可选（隐私优先）**：选本地则回复文本不出本机——
   - 本地 VITS（`sherpa-onnx-vits-zh-ll`，纯中文，5 说话人）；
-  - 本地 Kokoro（`kokoro-int8-multi-lang-v1_1`，**中英混读**，int8 约 109MB），经 `sherpa-onnx-node` **原生 addon** 运行（无 WASM 内存上限，连续合成不崩）；
+  - 本地 Kokoro（**中英混读**，103 音色；`int8` 默认约 109MB / 可选 `fp32` 约 311MB 音质更好），经 `sherpa-onnx-node` **原生 addon** 运行（无 WASM 内存上限，连续合成不崩）；
   - Edge 云端朗读保留为可选（设置 `ttsEngine: edge`）；设置面板「朗读引擎」热切换。
 - **Kokoro 音色全量 103 个**（F0 实测标定性别），四个常用男声置顶带编号；音色面板用 **下拉列表 + ◀▶ 步进**切换；
 - **增量传输**：partial 只传新增 0.9 秒，长段按住说话松手**秒出定稿**（不再整段重传重解码）；
@@ -73,6 +73,7 @@ bundle 插件安装后需重启 dsh 生效（Linux：`systemctl restart dsh`；
 | 键 | 默认 | 说明 |
 | --- | --- | --- |
 | `ttsEngine` | `edge` | 朗读引擎：`edge` 微软云端（默认，快）/ `vits` 本地中文 / `kokoro` 本地中英；**即时生效** |
+| `kokoroModel` | `int8` | Kokoro 模型精度：`int8`（默认，109MB，纯 CPU/低带宽推荐）/ `fp32`（311MB，音质更好，独显/大内存推荐）；两档共用 103 音色，**即时生效** |
 | `voice` | 按引擎 | 音色：VITS 五说话人；Kokoro 103 个（下拉+◀▶，62 深沉/68 浑厚/75 清亮/76 磁性置顶）；Edge 进入时自动加载全量 322 个。行内「试听」可即时预览 |
 | `rate` | `1.0` | 朗读语速倍率（0.5 慢速 ～ 2.0 快速），**即时生效** |
 | `interruptLevel` | `0` | 发声打断灵敏度（服务端 VAD 帧级检测 + 回声门控）：0 高门槛 / 1 中 / 2 低 |
@@ -82,9 +83,9 @@ bundle 插件安装后需重启 dsh 生效（Linux：`systemctl restart dsh`；
 | `autoSend` | `true` | 识别定稿后自动发送；关闭则只进草稿（按住 `Ctrl` / hold 松手仍会发送） |
 | `mode` | `toggle` | 交互模式：`toggle` 持续聆听 + 700ms 静音断句；`hold` 按住说话、松手发送（短按退出） |
 | `wakeWord` | 空（关） | 唤醒词（如「你好小D」）：进入后先说唤醒词激活，避免误触；空 = 关闭 |
-| `spokenFormat` | `false` | 语音会话注入口语化提示词：开启后**仅当前语音会话**的回复被注入「口语化短句、不用 Markdown 排版符号」提示词（朗读更顺），**即时生效** |
+| `spokenFormat` | `true` | 语音会话注入口语化提示词：开启后**仅当前语音会话**的回复被注入「口语化短句、不用 Markdown 排版符号」提示词（朗读更顺），**即时生效** |
 
-生效范围：`voice`/`rate`/`ttsEngine`/`spokenFormat` **立即生效**；其余设置下次进入语音模式时生效。设置项默认值由插件配置（`base` 层）提供。
+生效范围：`voice`/`rate`/`ttsEngine`/`kokoroModel`/`spokenFormat` **立即生效**；其余设置下次进入语音模式时生效。设置项默认值由插件配置（`base` 层）提供。
 
 ### 本地音色（VITS / Kokoro）
 

package/lib/client.js

@@ -984,7 +984,7 @@ var zh = {
   modeBtnToggle: "\u6301\u7EED\u8046\u542C",
   modeBtnHold: "\u6309\u4F4F\u8BF4",
   modeBtnTitle: "\u70B9\u51FB\u5207\u6362\u4EA4\u4E92\u6A21\u5F0F\uFF08\u4FDD\u5B58\u5230\u8BBE\u7F6E\uFF09",
-  descVoice: "Edge \u4E91\u7AEF\u97F3\u8272\uFF08\u8FDB\u5165\u65F6\u81EA\u52A8\u52A0\u8F7D\u5FAE\u8F6F\u5168\u90E8\u97F3\u8272\uFF0C\u4E0B\u62C9\u6216 \u25C0\u25B6 \u9009\uFF1B\u7F51\u7EDC\u5F02\u5E38\u56DE\u9000\u5E38\u7528 14 \u4E2A\uFF0C\u4E5F\u53EF\u300C\u81EA\u5B9A\u4E49\u300D\u586B ShortName\uFF09",
+  descVoice: "Edge \u4E91\u7AEF\u97F3\u8272\uFF08\u8FDB\u5165\u65F6\u81EA\u52A8\u52A0\u8F7D\u5FAE\u8F6F\u5168\u90E8\u97F3\u8272\uFF0C\u5E38\u7528\u4E2D\u6587\u97F3\u8272\u7F6E\u9876\uFF0C\u4E0B\u62C9\u6216 \u25C0\u25B6 \u9009\uFF1B\u4E5F\u53EF\u300C\u81EA\u5B9A\u4E49\u300D\u586B ShortName\uFF09",
   descVoiceLocal: "\u672C\u5730\u97F3\u8272\uFF08vits \u4E94\u4E2A\u8BF4\u8BDD\u4EBA\u5168\u90E8\u5217\u51FA\uFF0C\u4E0B\u62C9\u9009\u6216 \u25C0\u25B6 \u5207\u6362\uFF1B\u65E0\u9700\u81EA\u5B9A\u4E49\uFF09",
   ttsEngine: "\u6717\u8BFB\u5F15\u64CE",
   descTtsEngine: "\u672C\u5730 VITS\uFF08\u7EAF\u4E2D\u6587\uFF09/ \u672C\u5730 Kokoro\uFF08\u4E2D\u82F1\u5747\u53EF\uFF09/ Edge \u4E91\u7AEF\uFF08\u97F3\u8D28\u6700\u81EA\u7136\uFF0C\u6587\u672C\u4E0A\u5FAE\u8F6F\uFF09",
@@ -1020,7 +1020,7 @@ var zh = {
   descWakeWord: "\u5524\u9192\u8BCD\uFF08\u9ED8\u8BA4\u5173\uFF1B\u5982\u300C\u4F60\u597D\u5C0FD\u300D\uFF0C\u8BF4\u51FA\u540E\u5F00\u59CB\u8BC6\u522B\uFF09",
   wakePlaceholder: "\u5982\uFF1A\u4F60\u597D\u5C0FD",
   settingsCardDesc: "\u6717\u8BFB\u5F15\u64CE / \u97F3\u8272 / \u8BED\u901F / \u6253\u65AD\u7075\u654F\u5EA6 / \u6253\u65AD\u65B9\u5F0F / \u9759\u97F3\u505C\u987F / \u7A7A\u95F2\u8D85\u65F6 / \u6A21\u578B\u955C\u50CF / \u81EA\u52A8\u53D1\u9001 / \u4EA4\u4E92\u6A21\u5F0F / \u5524\u9192\u8BCD / \u53E3\u8BED\u5316\u63D0\u793A\u8BCD",
-  settingsEffectiveNote: "\u6717\u8BFB\u5F15\u64CE / \u97F3\u8272 / \u8BED\u901F / \u53E3\u8BED\u5316\u63D0\u793A\u8BCD / \u91CD\u8BD1 \u5373\u65F6\u751F\u6548\uFF1B\u5176\u4F59\uFF08\u6253\u65AD\u7075\u654F\u5EA6 / \u6253\u65AD\u65B9\u5F0F / \u56DE\u58F0\u95E8\u63A7 / \u5FEB\u6377\u952E / \u9759\u97F3 / \u7A7A\u95F2 / \u955C\u50CF / \u81EA\u52A8\u53D1\u9001 / \u81EA\u52A8\u6062\u590D / \u4EA4\u4E92\u6A21\u5F0F / \u5524\u9192\u8BCD / \u5DE5\u5177\u63D0\u793A\u97F3\uFF09\u4E0B\u6B21\u8FDB\u5165\u8BED\u97F3\u6A21\u5F0F\u65F6\u751F\u6548\u3002",
+  settingsEffectiveNote: "\u6717\u8BFB\u5F15\u64CE / \u97F3\u8272 / \u8BED\u901F / \u6A21\u578B\u7CBE\u5EA6 / \u53E3\u8BED\u5316\u63D0\u793A\u8BCD / \u91CD\u8BD1 \u5373\u65F6\u751F\u6548\uFF1B\u5176\u4F59\uFF08\u6253\u65AD\u7075\u654F\u5EA6 / \u6253\u65AD\u65B9\u5F0F / \u56DE\u58F0\u95E8\u63A7 / \u5FEB\u6377\u952E / \u9759\u97F3 / \u7A7A\u95F2 / \u955C\u50CF / \u81EA\u52A8\u53D1\u9001 / \u81EA\u52A8\u6062\u590D / \u4EA4\u4E92\u6A21\u5F0F / \u5524\u9192\u8BCD / \u5DE5\u5177\u63D0\u793A\u97F3\uFF09\u4E0B\u6B21\u8FDB\u5165\u8BED\u97F3\u6A21\u5F0F\u65F6\u751F\u6548\u3002",
   configUnavailable: "\u914D\u7F6E\u6682\u4E0D\u53EF\u7528",
   // telemetry（P1-5 开发模式延迟埋点状态条：各段耗时标签）
   telUtteranceEnd: "\u8BF4\u5B8C",
@@ -1053,6 +1053,16 @@ var zh = {
   ttsRedownload: "\u91CD\u65B0\u4E0B\u8F7D",
   ttsRedownloadHint: "\u6E05\u7406\u8BE5\u5F15\u64CE\u6A21\u578B\u7F13\u5B58\u5E76\u91CD\u65B0\u4E0B\u8F7D",
   ttsCleaning: "\u6E05\u7406\u4E2D\u2026",
+  ttsDownload: "\u4E0B\u8F7D",
+  ttsDelete: "\u5220\u9664",
+  ttsDownloading: "\u4E0B\u8F7D\u4E2D\u2026",
+  ttsDeleting: "\u5220\u9664\u4E2D\u2026",
+  ttsDownloadHint: "\u4E0B\u8F7D\u8BE5\u5F15\u64CE\u7684\u672C\u5730\u6A21\u578B\uFF1B\u4E0B\u8F7D\u5B8C\u6210\u540E\u7ACB\u5373\u5C31\u7EEA\uFF0C\u53EF\u8BD5\u542C/\u6717\u8BFB",
+  ttsDeleteHint: "\u5220\u9664\u672C\u5730\u6A21\u578B\uFF08\u91CA\u653E\u7A7A\u95F4\uFF1B\u4E0B\u6B21\u4F7F\u7528\u4F1A\u81EA\u52A8\u91CD\u65B0\u4E0B\u8F7D\uFF09",
+  kokoroModel: "Kokoro \u6A21\u578B\u7CBE\u5EA6",
+  kokoroModelInt8: "int8\uFF08\u9ED8\u8BA4\uFF09",
+  kokoroModelFp32: "fp32\uFF08\u97F3\u8D28\u66F4\u597D\uFF09",
+  descKokoroModel: "Kokoro \u6A21\u578B\u7CBE\u5EA6\uFF1Aint8 \u4F53\u79EF\u5C0F/\u52A0\u8F7D\u5FEB\uFF08\u7EAF CPU \u670D\u52A1\u5668\u6216\u4F4E\u5E26\u5BBD\u63A8\u8350\uFF0C\u9ED8\u8BA4\uFF09\uFF1Bfp32 \u97F3\u8D28\u66F4\u597D\u4F46\u7EA6 311MB\u3001\u66F4\u6162\uFF08\u6709\u72EC\u7ACB\u663E\u5361\u6216\u5927\u5185\u5B58\u673A\u5668\u63A8\u8350\uFF09\u3002\u4E24\u6863\u5171\u7528\u540C\u4E00\u5957 103 \u97F3\u8272\uFF0C\u5207\u6362\u5373\u65F6\u751F\u6548\u3002",
   // 设置分组标题
   secRead: "\u6717\u8BFB\u4E0E\u97F3\u8272",
   secInterrupt: "\u6253\u65AD\u4E0E\u9759\u97F3",
@@ -1115,7 +1125,7 @@ var en = {
   modeBtnToggle: "Continuous",
   modeBtnHold: "Hold to talk",
   modeBtnTitle: "Click to switch interaction mode (saved to settings)",
-  descVoice: "Edge cloud voices (auto-loads all Microsoft voices on entry; \u25C0\u25B6 or dropdown; falls back to 14 common on network failure; custom ShortName allowed)",
+  descVoice: "Edge cloud voices (auto-loads all Microsoft voices; common Chinese voices pinned on top; \u25C0\u25B6 or dropdown; custom ShortName allowed)",
   descVoiceLocal: "Local voice (vits, all 5 speakers listed; dropdown or \u25C0\u25B6; no custom needed)",
   ttsEngine: "Read-aloud engine",
   descTtsEngine: "Local VITS (Chinese only) / Local Kokoro (Chinese + English) / Edge cloud (most natural, text sent to Microsoft)",
@@ -1151,7 +1161,7 @@ var en = {
   descWakeWord: "Wake word (default off; e.g. Hey D)",
   wakePlaceholder: "e.g. Hey D",
   settingsCardDesc: "Engine / voice / rate / interrupt / barge-in / silence / idle / model host / auto-send / mode / wake word / spoken format",
-  settingsEffectiveNote: "Engine / voice / rate / spoken format / re-transcribe apply immediately; the rest (interrupt / barge-in / echo gate / shortcut / silence / idle / mirror / auto-send / auto-resume / mode / wake word / tool beep) apply next time you enter voice mode.",
+  settingsEffectiveNote: "Engine / voice / rate / model precision / spoken format / re-transcribe apply immediately; the rest (interrupt / barge-in / echo gate / shortcut / silence / idle / mirror / auto-send / auto-resume / mode / wake word / tool beep) apply next time you enter voice mode.",
   configUnavailable: "Configuration unavailable",
   telUtteranceEnd: "end",
   telEndpoint: "endpoint",
@@ -1181,6 +1191,16 @@ var en = {
   ttsRedownload: "Re-download",
   ttsRedownloadHint: "Clear this engine model cache and re-download",
   ttsCleaning: "Clearing\u2026",
+  ttsDownload: "Download",
+  ttsDelete: "Delete",
+  ttsDownloading: "Downloading\u2026",
+  ttsDeleting: "Deleting\u2026",
+  ttsDownloadHint: "Download this engine's local model; becomes ready immediately after",
+  ttsDeleteHint: "Delete local models (frees space; auto re-downloads on next use)",
+  kokoroModel: "Kokoro model precision",
+  kokoroModelInt8: "int8 (default)",
+  kokoroModelFp32: "fp32 (better quality)",
+  descKokoroModel: "Kokoro model precision: int8 is smaller/faster (CPU server or low bandwidth; default); fp32 sounds better but ~311MB and slower (GPU or large memory). Both share the same 103 voices; switches live.",
   secRead: "Reading & voice",
   secInterrupt: "Interrupt & silence",
   secInteraction: "Interaction",
@@ -1213,6 +1233,7 @@ var cardStyle = {
 };
 var FIELD_LABELS = {
   ttsEngine: "\u6717\u8BFB\u5F15\u64CE",
+  kokoroModel: "Kokoro \u6A21\u578B\u7CBE\u5EA6",
   voice: "\u97F3\u8272",
   rate: "\u8BED\u901F",
   interruptLevel: "\u6253\u65AD\u7075\u654F\u5EA6",
@@ -1445,8 +1466,10 @@ var VOICE_OPTIONS_KOKORO = [
   ...KOKORO_F0.map((_, sid) => kokoroOption(sid)).filter((o) => !KOKORO_PINNED.includes(Number(o.v)))
 ];
 var ENGINE_DEFAULT_VOICE = {
-  vits: "fushiyu",
-  kokoro: "zf_xiaoni",
+  // 与 host 侧引擎 defaultVoice 对齐（VITS suyingxue / Kokoro zf_xiaobei），
+  // 避免「config 直连」与「面板切引擎」落到不同默认音色。
+  vits: "suyingxue",
+  kokoro: "zf_xiaobei",
   edge: "zh-CN-XiaoxiaoNeural"
 };
 var HOST_OPTIONS = [
@@ -1619,6 +1642,10 @@ function VoiceSelect({
   };
   const selectStyle = {
     ...inputStyle,
+    // 下拉主控件占满剩余宽度（覆盖 inputStyle 固定 280，避免与 ‹› 并排时被压窄导致长名截断）。
+    width: "auto",
+    minWidth: 0,
+    flex: "1 1 auto",
     appearance: "none",
     cursor: "pointer",
     backgroundImage: `url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='12' viewBox='0 0 12 12' fill='none'%3E%3Cpath d='M3 4.5L6 7.5L9 4.5' stroke='%2381858C' stroke-width='1.5' stroke-linecap='round' stroke-linejoin='round'/%3E%3C/svg%3E")`,
@@ -1629,7 +1656,7 @@ function VoiceSelect({
   };
   const label = inOptions ? options[idx].label : custom || placeholder || "";
   const selectValue = inOptions ? cur : showCustom ? "__custom__" : options[0]?.v ?? "";
-  return /* @__PURE__ */ (0, import_jsx_runtime.jsxs)("span", { style: { display: "flex", flexDirection: "column", gap: 6, width: 320, maxWidth: "100%", alignItems: "stretch" }, children: [
+  return /* @__PURE__ */ (0, import_jsx_runtime.jsxs)("span", { style: { display: "flex", flexDirection: "column", gap: 6, width: "100%", maxWidth: "100%", alignItems: "stretch" }, children: [
     /* @__PURE__ */ (0, import_jsx_runtime.jsxs)("span", { style: { display: "flex", gap: 6, alignItems: "center" }, children: [
       /* @__PURE__ */ (0, import_jsx_runtime.jsx)("button", { type: "button", "aria-label": t("voicePrev"), onClick: () => move(-1), style: stepBtn, children: "\u2039" }),
       /* @__PURE__ */ (0, import_jsx_runtime.jsxs)(
@@ -1805,7 +1832,7 @@ function SegGroup({
 var fmtMB = (b) => b >= 1048576 ? `${(b / 1048576).toFixed(0)}MB` : b > 0 ? `${Math.round(b / 1024)}KB` : "\u2013";
 function EngineStatusInline() {
   const [st, setSt] = (0, import_react.useState)(null);
-  const [cleaning, setCleaning] = (0, import_react.useState)(false);
+  const [acting, setActing] = (0, import_react.useState)(null);
   (0, import_react.useEffect)(() => {
     let alive = true;
     const poll = async () => {
@@ -1822,51 +1849,50 @@ function EngineStatusInline() {
       clearInterval(timer);
     };
   }, []);
-  const clean = (engine) => {
-    setCleaning(true);
-    void fetch(location.origin + BASE_PATH + "/models/clean", {
-      method: "POST",
-      headers: { "content-type": "application/json" },
-      body: JSON.stringify({ engine })
-    }).catch(() => void 0).finally(() => setTimeout(() => setCleaning(false), 1500));
-  };
   const tts = st?.tts;
   if (!tts) return /* @__PURE__ */ (0, import_jsx_runtime.jsx)(import_jsx_runtime.Fragment, {});
   const engineName = tts.engine === "vits" ? t("engineVits") : tts.engine === "kokoro" ? t("engineKokoro") : t("engineEdge");
   const isLocal = !!tts.local;
+  const localReady = isLocal ? !!tts.local?.ready : false;
   let statusText;
   let statusColor;
   if (tts.loading) {
     statusText = t("engineLoading");
     statusColor = t2.term;
-  } else if (tts.ready) {
-    statusText = t("engineReady");
-    statusColor = "var(--dsw-alias-state-success-primary)";
   } else if (tts.error) {
     statusText = t("engineError");
     statusColor = "var(--dsw-alias-state-error-primary)";
-  } else if (isLocal && !tts.local.ready) {
+  } else if (isLocal && !localReady) {
     statusText = t("ttsModelsMissing");
     statusColor = "var(--dsw-alias-state-error-primary)";
   } else {
-    statusText = t("engineIdle");
-    statusColor = t2.term;
+    statusText = t("engineReady");
+    statusColor = "var(--dsw-alias-state-success-primary)";
   }
+  const act = (kind) => {
+    setActing(kind);
+    void fetch(location.origin + BASE_PATH + (kind === "clean" ? "/models/clean" : "/models/download"), {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({ engine: tts.engine })
+    }).catch(() => void 0).finally(() => setTimeout(() => setActing(null), 1500));
+  };
+  const action = localReady ? "clean" : "download";
   return /* @__PURE__ */ (0, import_jsx_runtime.jsxs)("div", { style: { display: "flex", alignItems: "center", flexWrap: "wrap", gap: 8, padding: "2px 0 10px" }, children: [
     /* @__PURE__ */ (0, import_jsx_runtime.jsx)("span", { style: { fontSize: 12, color: t2.term }, children: engineName }),
-    /* @__PURE__ */ (0, import_jsx_runtime.jsx)("span", { style: { fontSize: 12, fontWeight: tts.ready || tts.error ? 600 : 400, color: statusColor }, children: statusText }),
-    tts.loading && st?.progress?.file && /* @__PURE__ */ (0, import_jsx_runtime.jsxs)("span", { style: { fontSize: 12, color: t2.term }, children: [
-      st.progress.file,
+    /* @__PURE__ */ (0, import_jsx_runtime.jsx)("span", { style: { fontSize: 12, fontWeight: statusText === t("engineReady") || statusText === t("engineError") ? 600 : 400, color: statusColor }, children: statusText }),
+    tts.loading && tts.progress?.file && /* @__PURE__ */ (0, import_jsx_runtime.jsxs)("span", { style: { fontSize: 12, color: t2.term }, children: [
+      tts.progress.file,
       " ",
-      st.progress.percent,
+      tts.progress.percent,
       "%"
     ] }),
     isLocal && /* @__PURE__ */ (0, import_jsx_runtime.jsx)(
       "button",
       {
         type: "button",
-        onClick: () => clean(tts.engine),
-        disabled: cleaning || tts.loading,
+        onClick: () => act(action),
+        disabled: !!acting || tts.loading,
         style: {
           font: "inherit",
           fontSize: 12,
@@ -1878,8 +1904,8 @@ function EngineStatusInline() {
           padding: "3px 10px",
           flexShrink: 0
         },
-        title: t("ttsRedownloadHint"),
-        children: cleaning ? t("ttsCleaning") : t("ttsRedownload")
+        title: action === "clean" ? t("ttsDeleteHint") : t("ttsDownloadHint"),
+        children: acting ? acting === "clean" ? t("ttsDeleting") : t("ttsDownloading") : action === "clean" ? t("ttsDelete") : t("ttsDownload")
       }
     ),
     tts.error && /* @__PURE__ */ (0, import_jsx_runtime.jsx)("span", { style: { fontSize: 11, color: "var(--dsw-alias-state-error-primary)", flexBasis: "100%" }, children: tts.error })
@@ -1996,17 +2022,16 @@ function VoiceSettingsCard({ scope }) {
     let alive = true;
     void fetch(location.origin + BASE_PATH + "/voices").then((res) => res.ok ? res.json() : null).then((data) => {
       if (!alive || !data?.voices) return;
-      const localeName = (lk) => {
-        const m = /^([a-z]{2,3})(?:-[A-Z]{2})?/.exec(lk);
-        return m ? m[1].toUpperCase() : lk;
-      };
       const genderName = (g) => g === "Female" ? "\u5973" : g === "Male" ? "\u7537" : g === "Neutral" ? "\u4E2D\u6027" : g;
-      setEdgeVoices(
-        data.voices.map((v) => ({
-          v: v.ShortName,
-          label: v.FriendlyName.replace(/ Microsoft.*/, "") + " \xB7 " + genderName(v.Gender) + " \xB7 " + localeName(v.Locale) + " (" + v.ShortName + ")"
-        })).sort((a, b) => a.label.localeCompare(b.label))
+      const voiceName = (fn) => fn.replace(/^Microsoft\s+/, "").replace(/\s+Online\s+\(Natural\)/, "").split(/\s*-\s*/)[0].trim();
+      const mkLabel = (v) => (
+        // 精简标签：只留说话人名 + 性别（去掉尾部区域与 ShortName 冗余），避免 Edge 长名被截断。
+        (voiceName(v.FriendlyName) || v.ShortName) + " \xB7 " + genderName(v.Gender)
       );
+      const all = data.voices.map((v) => ({ v: v.ShortName, label: mkLabel(v) })).sort((a, b) => a.label.localeCompare(b.label));
+      const commonKeys = new Set(VOICE_OPTIONS.map((o) => o.v));
+      const pinned = VOICE_OPTIONS.filter((o) => all.some((a) => a.v === o.v));
+      setEdgeVoices([...pinned, ...all.filter((a) => !commonKeys.has(a.v))]);
     }).catch(() => void 0);
     return () => {
       alive = false;
@@ -2049,6 +2074,18 @@ function VoiceSettingsCard({ scope }) {
           }
         ) }),
         /* @__PURE__ */ (0, import_jsx_runtime.jsx)(EngineStatusInline, {}),
+        engine === "kokoro" && /* @__PURE__ */ (0, import_jsx_runtime.jsx)(Row, { name: "kokoroModel", desc: t("descKokoroModel"), children: /* @__PURE__ */ (0, import_jsx_runtime.jsx)(
+          SegGroup,
+          {
+            score: scope,
+            field: "kokoroModel",
+            value: value.kokoroModel,
+            options: [
+              { v: "int8", label: t("kokoroModelInt8") },
+              { v: "fp32", label: t("kokoroModelFp32") }
+            ]
+          }
+        ) }),
         /* @__PURE__ */ (0, import_jsx_runtime.jsx)(
           Row,
           {
@@ -2147,7 +2184,7 @@ var TELEMETRY_VIEW = [
   { stage: "first-tts-chunk", key: "telFirstChunk" },
   { stage: "first-audio-played", key: "telFirstPlayed" }
 ];
-var BUILD_TAG = "a76aad8";
+var BUILD_TAG = "d4c61da";
 var TELEMETRY_FLAG = "dsh-voice-mode.telemetry";
 var telemetryEnabled = typeof localStorage !== "undefined" && localStorage.getItem(TELEMETRY_FLAG) === "1";
 console.log("[dsh-voice] build=" + BUILD_TAG);
@@ -3245,7 +3282,11 @@ function MicButton({
       if (sid2) void bus.exit(sid2);
     }
   };
+  const toggleGuardRef = (0, import_react2.useRef)(0);
   const toggle = () => {
+    const now = Date.now();
+    if (now - toggleGuardRef.current < 2e3) return;
+    toggleGuardRef.current = now;
     if (localRef.current === "on") void exitModeRef.current("manual");
     else if (localRef.current === "off") void enterMode();
   };
@@ -3520,6 +3561,9 @@ function MicButton({
       return;
     }
     if (ms < 250) {
+      const now = Date.now();
+      if (now - toggleGuardRef.current < 2e3) return;
+      toggleGuardRef.current = now;
       if (localRef.current === "on") {
         engineRef.current?.endHeld(true);
         void exitModeRef.current("manual");

package/lib/index.js

@@ -1129,6 +1129,10 @@ var TtsQueue = class {
     if (s) return s;
     return { engine: "edge", ready: true, loading: false };
   }
+  /** 触发当前引擎预热/下载模型并初始化（设置面板「下载」按钮；Edge 为无操作）。 */
+  async prepare() {
+    await this.engine.prepare?.();
+  }
   /**
    * 一次性合成（设置卡「试听」用）：委托当前引擎；不干扰朗读队列的在途合成。
    * 失败（含非法音色）抛错。
@@ -1248,7 +1252,9 @@ import { fileURLToPath as fileURLToPath2 } from "node:url";
 import { join as join3 } from "node:path";
 import { statSync as statSync2 } from "node:fs";
 var TTS_MODEL_REPO = "csukuangfj/sherpa-onnx-vits-zh-ll";
-var KOKORO_MODEL_DIR = "csukuangfj/kokoro-int8-multi-lang-v1_1";
+var KOKORO_MODEL_DIR_INT8 = "csukuangfj/kokoro-int8-multi-lang-v1_1";
+var KOKORO_MODEL_DIR_FP32 = "csukuangfj/kokoro-multi-lang-v1_1";
+var kokoroModelDir = (m) => m === "fp32" ? KOKORO_MODEL_DIR_FP32 : KOKORO_MODEL_DIR_INT8;
 var TTS_MODEL_FILES = [
   { file: "model.onnx", sha256: "6c349bdd73dc928234dd7bc86929748bba32cd5264d32d915bf7b7aa0595965b" },
   { file: "lexicon.txt", sha256: "b3a82f16b286c424953dea3686039e7ab465fa8e15d87ef8abd0ec69175beb21" },
@@ -1452,36 +1458,51 @@ var VITS_SPEC = {
   defaultVoice: "suyingxue",
   toSid: voiceToSid
 };
-var KOKORO_SPEC = {
-  files: [
-    { file: "model.int8.onnx", sha256: "bda15858163726a492d02a9a727bc263551b86ac77f90812c4b30ff41d380e26" },
-    { file: "voices.bin", sha256: "e64a5a581d8c2a350d848f51c3121657cd83aa07ed6109172177345874a7244c" },
-    { file: "tokens.txt", sha256: "931ab2df2400cd65d580a22402024c2347ced8ae9ea300e545144b1aacc48e14" },
-    { file: "lexicon-us-en.txt", sha256: "7daaab53a181be9885b853a8582bf1838186317e5dadacbcef9c426d6fa0da14" },
-    { file: "lexicon-zh.txt", sha256: "11111d8cd695fba2ace1367a1d0a708b586e6ef5c1f9be91da5d7eef129b651c" },
-    { file: "espeak-ng-data/phontab", sha256: "886f3fa402cb0ba73d483aa8ad000af47a6b7cc06293c75a97913fba68a530f6" },
-    { file: "date-zh.fst", sha256: "eb8aa079ae3cb81d8f4404992f39d61a0cb990947512b5b8d1e54d1f6980e718" },
-    { file: "number-zh.fst", sha256: "743f402181fcfebf76cc2f0546b71fa26476e626fbe4e460fb7b4c3a7a8bd5bd" },
-    { file: "phone-zh.fst", sha256: "1ac2b6fa56b1442320c4de7db08353bab8963a2b57f365eebcdd3a2d3562f8d7" }
-  ],
-  workerPaths: (dir) => ({
-    model: join3(dir, "model.int8.onnx"),
-    voices: join3(dir, "voices.bin"),
-    tokens: join3(dir, "tokens.txt"),
-    dataDir: join3(dir, "espeak-ng-data"),
-    lexicon: [join3(dir, "lexicon-us-en.txt"), join3(dir, "lexicon-zh.txt")].join(","),
-    date: join3(dir, "date-zh.fst"),
-    phone: join3(dir, "phone-zh.fst"),
-    number: join3(dir, "number-zh.fst"),
-    lang: ""
-  }),
-  defaultVoice: "zf_xiaobei",
-  toSid: kokoroVoiceToSid
-};
+var KOKORO_SHARED_FILES = [
+  { file: "voices.bin", sha256: "e64a5a581d8c2a350d848f51c3121657cd83aa07ed6109172177345874a7244c" },
+  { file: "tokens.txt", sha256: "931ab2df2400cd65d580a22402024c2347ced8ae9ea300e545144b1aacc48e14" },
+  { file: "lexicon-us-en.txt", sha256: "7daaab53a181be9885b853a8582bf1838186317e5dadacbcef9c426d6fa0da14" },
+  { file: "lexicon-zh.txt", sha256: "11111d8cd695fba2ace1367a1d0a708b586e6ef5c1f9be91da5d7eef129b651c" },
+  { file: "espeak-ng-data/phontab", sha256: "886f3fa402cb0ba73d483aa8ad000af47a6b7cc06293c75a97913fba68a530f6" },
+  { file: "date-zh.fst", sha256: "eb8aa079ae3cb81d8f4404992f39d61a0cb990947512b5b8d1e54d1f6980e718" },
+  { file: "number-zh.fst", sha256: "743f402181fcfebf76cc2f0546b71fa26476e626fbe4e460fb7b4c3a7a8bd5bd" },
+  { file: "phone-zh.fst", sha256: "1ac2b6fa56b1442320c4de7db08353bab8963a2b57f365eebcdd3a2d3562f8d7" }
+];
+var kokoroWorkerPaths = (dir, modelFile) => ({
+  model: join3(dir, modelFile),
+  voices: join3(dir, "voices.bin"),
+  tokens: join3(dir, "tokens.txt"),
+  dataDir: join3(dir, "espeak-ng-data"),
+  lexicon: [join3(dir, "lexicon-us-en.txt"), join3(dir, "lexicon-zh.txt")].join(","),
+  date: join3(dir, "date-zh.fst"),
+  phone: join3(dir, "phone-zh.fst"),
+  number: join3(dir, "number-zh.fst"),
+  lang: ""
+});
+function kokoroSpec(model) {
+  const main = model === "fp32" ? { file: "model.onnx", sha256: "acc4adc175b9d9986106cd20060329673ad5a2e12ef3c557d2d3745b694f8b38" } : { file: "model.int8.onnx", sha256: "bda15858163726a492d02a9a727bc263551b86ac77f90812c4b30ff41d380e26" };
+  return {
+    files: [main, ...KOKORO_SHARED_FILES],
+    workerPaths: (dir) => kokoroWorkerPaths(dir, main.file),
+    defaultVoice: "zf_xiaobei",
+    toSid: kokoroVoiceToSid
+  };
+}
 function createSherpaLocalEngine(options) {
   const { cacheDir, modelHost, allowCustomHost, broadcast } = options;
-  const spec = options.kind === "kokoro" ? KOKORO_SPEC : VITS_SPEC;
-  const repoName = options.kind === "kokoro" ? KOKORO_MODEL_DIR : TTS_MODEL_REPO;
+  let downloadProgress = null;
+  const trackedBroadcast = (event, payload) => {
+    if (event === "asr-progress" && payload && typeof payload === "object") {
+      const pr = payload;
+      if (typeof pr.file === "string" && typeof pr.percent === "number") {
+        downloadProgress = { file: pr.file, percent: Math.min(100, Math.max(0, pr.percent)) };
+      }
+    }
+    broadcast(event, payload);
+  };
+  const kokoroModel = options.model ?? "int8";
+  const spec = options.kind === "kokoro" ? kokoroSpec(kokoroModel) : VITS_SPEC;
+  const repoName = options.kind === "kokoro" ? kokoroModelDir(kokoroModel) : TTS_MODEL_REPO;
   const repoDir = join3(cacheDir, repoName);
   const workerPath = fileURLToPath2(new URL("./tts-vits-worker.cjs", import.meta.url));
   let child = null;
@@ -1520,6 +1541,7 @@ function createSherpaLocalEngine(options) {
       ready = (async () => {
         engineLoading = true;
         engineError = void 0;
+        downloadProgress = null;
         try {
           if (!child || !childInit) {
             for (const f of spec.files) {
@@ -1529,7 +1551,7 @@ function createSherpaLocalEngine(options) {
                 spec: f,
                 primaryHost: modelHost(),
                 allowCustomHost,
-                broadcast
+                broadcast: trackedBroadcast
               });
               if (!ok) throw new Error("local TTS model download/verify failed: " + f.file);
             }
@@ -1540,7 +1562,7 @@ function createSherpaLocalEngine(options) {
                 subdir: "espeak-ng-data",
                 primaryHost: modelHost(),
                 allowCustomHost,
-                broadcast
+                broadcast: trackedBroadcast
               });
               if (!treeOk) throw new Error("local TTS model download failed: espeak-ng-data");
             }
@@ -1623,6 +1645,7 @@ function createSherpaLocalEngine(options) {
         ready: childInit === true,
         loading: engineLoading,
         error: engineError,
+        progress: downloadProgress ?? void 0,
         local: {
           repo: repoName,
           ready: files.every((f) => f.exists),
@@ -1632,6 +1655,8 @@ function createSherpaLocalEngine(options) {
         }
       };
     },
+    // 设置面板「下载」按钮：无文本也触发模型下载 + 子进程初始化（与首次合成路径一致）。
+    prepare: () => ensureReady(),
     async synthesize(text, opts = {}) {
       await ensureReady();
       const sid = spec.toSid(opts.voice ?? voice);
@@ -1744,6 +1769,7 @@ var inject = ["webServer", "settings", "sessions"];
 var defaultModelCacheDir = () => process.platform === "win32" ? join4(process.env.LOCALAPPDATA ?? join4(homedir(), "AppData", "Local"), "dsh-voice-mode", "models") : join4(homedir(), ".cache", "dsh-voice-mode", "models");
 var VOICE_SETTINGS_DEFAULTS = {
   ttsEngine: "edge",
+  kokoroModel: "int8",
   voice: "zh-CN-XiaoxiaoNeural",
   rate: 1,
   interruptLevel: 0,
@@ -1767,6 +1793,9 @@ function createVoiceSettingsSchema(defs) {
     ttsEngine: z.union([z.const("vits"), z.const("kokoro"), z.const("edge")]).default(d.ttsEngine).description(
       "\u6717\u8BFB\u5F15\u64CE\uFF1Aedge \u5FAE\u8F6F\u4E91\u7AEF\uFF08\u9ED8\u8BA4\uFF0C\u5FEB\u3001\u97F3\u8D28\u81EA\u7136\uFF0C\u88AB\u6717\u8BFB\u6587\u672C\u4F1A\u53D1\u9001\u5230\u5FAE\u8F6F\uFF09/ vits \u672C\u5730\u4E2D\u6587 / kokoro \u672C\u5730\u4E2D\u82F1\uFF08\u56DE\u590D\u6587\u672C\u4E0D\u51FA\u672C\u673A\uFF09\uFF1B\u5207\u6362\u5373\u65F6\u751F\u6548"
     ),
+    kokoroModel: z.union([z.const("int8"), z.const("fp32")]).default(d.kokoroModel).description(
+      "Kokoro \u6A21\u578B\u7CBE\u5EA6\uFF1Aint8\uFF08\u9ED8\u8BA4\uFF0C\u4F53\u79EF\u5C0F/\u52A0\u8F7D\u5FEB\uFF0CCPU \u53CB\u597D\uFF09/ fp32\uFF08\u97F3\u8D28\u66F4\u597D\u3001\u4F53\u79EF\u5927\uFF0CGPU \u6216\u5927\u5185\u5B58\u673A\u5668\u63A8\u8350\uFF09\uFF1B\u4E24\u6863\u5171\u7528\u540C\u4E00\u5957 103 \u97F3\u8272\uFF0C\u5207\u6362\u5373\u65F6\u751F\u6548"
+    ),
     voice: z.string().default(d.voice).description(
       "\u6717\u8BFB\u97F3\u8272\uFF08\u6309 ttsEngine \u53D6\u503C\uFF1Avits \u7528\u8BF4\u8BDD\u4EBA\u540D suyingxue/gunian/fushiyu/bingjiao/bazong\uFF1Bkokoro \u7528 0-102 \u7F16\u53F7\u6216\u4E2D\u6587\u540D zf_xiaobei/zf_xiaoni/zf_xiaoxiao/zf_xiaoyi\uFF1Bedge \u7528 Edge ShortName \u5982 zh-CN-XiaoxiaoNeural \u6653\u6653\xB7\u5973\uFF0C\u5B8C\u6574\u6E05\u5355\u89C1 scripts/list-voices.mjs\uFF09"
     ),
@@ -1793,6 +1822,7 @@ var Config = z.object({
   cacheDir: z.string().default(defaultModelCacheDir()),
   modelHost: z.string().default("https://huggingface.co"),
   ttsEngine: z.union([z.const("edge"), z.const("vits"), z.const("kokoro")]).default("edge"),
+  kokoroModel: z.union([z.const("int8"), z.const("fp32")]).default("int8"),
   allowLan: z.boolean().default(false),
   allowCustomModelHost: z.boolean().default(false),
   voice: z.string().default("zh-CN-XiaoxiaoNeural"),
@@ -1878,6 +1908,7 @@ function apply(ctx, config) {
         cacheDir: config.cacheDir,
         modelHost: normalizedModelHost,
         allowCustomHost: config.allowCustomModelHost,
+        model: vset.kokoroModel,
         broadcast
       });
     }
@@ -1889,6 +1920,7 @@ function apply(ctx, config) {
     });
   };
   let engineKind = vset.ttsEngine ?? config.ttsEngine;
+  let activeKokoroModel = vset.kokoroModel;
   const queue = new TtsQueue({
     engine: makeEngine(engineKind),
     onError: (sessionId) => broadcast("tts-error", { sessionId })
@@ -1903,6 +1935,9 @@ function apply(ctx, config) {
       if (next.ttsEngine !== engineKind) {
         engineKind = next.ttsEngine;
         queue.setEngine(makeEngine(engineKind));
+      } else if (engineKind === "kokoro" && next.kokoroModel !== activeKokoroModel) {
+        activeKokoroModel = next.kokoroModel;
+        queue.setEngine(makeEngine("kokoro"));
       }
       queue.updateVoice(next.voice, next.rate);
     })
@@ -2072,7 +2107,7 @@ function apply(ctx, config) {
             respondJson2(res, 400, { error: "invalid on" });
             return;
           }
-          if (!limiter.hit(`toggle:${sessionId}`, 1, 2e3)) {
+          if (!limiter.hit(`toggle:${sessionId}`, 2, 2e3)) {
             res.statusCode = 429;
             res.setHeader("content-type", "application/json");
             res.end(JSON.stringify({ error: "rate limited" }));
@@ -2189,7 +2224,7 @@ function apply(ctx, config) {
             respondJson2(res, 400, { error: "invalid json" });
             return;
           }
-          const dir = join4(config.cacheDir, engine === "kokoro" ? KOKORO_MODEL_DIR : TTS_MODEL_REPO);
+          const dir = join4(config.cacheDir, engine === "kokoro" ? kokoroModelDir(vset.kokoroModel) : TTS_MODEL_REPO);
           void rm(dir, { recursive: true, force: true }).then(() => {
             if (engineKind === engine) {
               queue.setEngine(makeEngine(engine));
@@ -2201,6 +2236,42 @@ function apply(ctx, config) {
       }
     })
   );
+  ctx.effect(
+    () => ctx.webServer.register({
+      kind: "exact",
+      path: `${base}/models/download`,
+      handler: (req, res) => {
+        if (denyNonLoopback(req, res)) return;
+        if (denyCrossOrigin(req, res)) return;
+        if (!config.enabled) {
+          respondJson2(res, 403, { error: "voice mode disabled" });
+          return;
+        }
+        collectBody(req, res, MAX_JSON_BODY, (body) => {
+          let engine = "vits";
+          try {
+            const p = JSON.parse(body || "{}");
+            if (p.engine === "kokoro" || p.engine === "vits") engine = p.engine;
+            else {
+              respondJson2(res, 400, { error: "invalid engine" });
+              return;
+            }
+          } catch {
+            respondJson2(res, 400, { error: "invalid json" });
+            return;
+          }
+          if (engineKind !== engine) {
+            respondJson2(res, 400, { error: "engine not active" });
+            return;
+          }
+          void queue.prepare().then(() => respondJson2(res, 200, { ok: true, engine })).catch((e) => {
+            console.warn(`[dsh-voice-mode] model download failed: ${String(e)}`);
+            respondJson2(res, 502, { error: "\u6A21\u578B\u4E0B\u8F7D\u5931\u8D25\uFF1A\u8BF7\u68C0\u67E5\u7F51\u7EDC" });
+          });
+        });
+      }
+    })
+  );
   ctx.effect(
     () => ctx.webServer.register({
       kind: "exact",

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "dsh-voice-mode",
   "description": "Full-duplex voice mode for DeepSeek Harness: zipformer2 streaming ASR → editable draft, Edge TTS (default, fast) or local (VITS / Kokoro int8, privacy-first) read-aloud with live captions, true barge-in; hardened HTTP surface (loopback/origin/rate-limit) + model SHA256 pinning. · DSH 语音双工对话：流式识别入草稿、Edge 默认朗读或本地（VITS / Kokoro int8）朗读+实时字幕、开口即打断；HTTP 面安全加固。",
-  "version": "0.5.0",
+  "version": "0.6.0",
   "type": "module",
   "main": "lib/index.js",
   "repository": {

package/scripts/prefetch.mjs

@@ -51,6 +51,18 @@ const MODELS = {
     'number-zh.fst': '743f402181fcfebf76cc2f0546b71fa26476e626fbe4e460fb7b4c3a7a8bd5bd',
     'phone-zh.fst': '1ac2b6fa56b1442320c4de7db08353bab8963a2b57f365eebcdd3a2d3562f8d7',
   },
+  // 本地 TTS：Kokoro fp32（音质更好，约 311MB；支持文件与 int8 同源）
+  'csukuangfj/kokoro-multi-lang-v1_1': {
+    'model.onnx': 'acc4adc175b9d9986106cd20060329673ad5a2e12ef3c557d2d3745b694f8b38',
+    'voices.bin': 'e64a5a581d8c2a350d848f51c3121657cd83aa07ed6109172177345874a7244c',
+    'tokens.txt': '931ab2df2400cd65d580a22402024c2347ced8ae9ea300e545144b1aacc48e14',
+    'lexicon-us-en.txt': '7daaab53a181be9885b853a8582bf1838186317e5dadacbcef9c426d6fa0da14',
+    'lexicon-zh.txt': '11111d8cd695fba2ace1367a1d0a708b586e6ef5c1f9be91da5d7eef129b651c',
+    'espeak-ng-data/phontab': '886f3fa402cb0ba73d483aa8ad000af47a6b7cc06293c75a97913fba68a530f6',
+    'date-zh.fst': 'eb8aa079ae3cb81d8f4404992f39d61a0cb990947512b5b8d1e54d1f6980e718',
+    'number-zh.fst': '743f402181fcfebf76cc2f0546b71fa26476e626fbe4e460fb7b4c3a7a8bd5bd',
+    'phone-zh.fst': '1ac2b6fa56b1442320c4de7db08353bab8963a2b57f365eebcdd3a2d3562f8d7',
+  },
 }
 const HOST_PRIMARY = 'https://huggingface.co'
 const HOST_FALLBACK = 'https://hf-mirror.com'
@@ -150,21 +162,25 @@ for (const [repo, files] of Object.entries(MODELS)) {
   }
 }
 
-// --- Kokoro 中英朗读模型：文件多（含 espeak-ng-data 全目录），走 HF 树枚举批量下载 ---
-{
-  const repo = 'csukuangfj/kokoro-int8-multi-lang-v1_1'
+// --- Kokoro 中英朗读模型：文件多（含 espeak-ng-data 全目录），走 HF 树枚举批量下载。
+//     int8（默认，CPU 友好）与 fp32（音质更好）两档，主模型文件不同、支持文件同源。 ---
+for (const kokoro of [
+  { repo: 'csukuangfj/kokoro-int8-multi-lang-v1_1', main: 'model.int8.onnx', sha: 'bda15858163726a492d02a9a727bc263551b86ac77f90812c4b30ff41d380e26' },
+  { repo: 'csukuangfj/kokoro-multi-lang-v1_1', main: 'model.onnx', sha: 'acc4adc175b9d9986106cd20060329673ad5a2e12ef3c557d2d3745b694f8b38' },
+]) {
+  const { repo, main, sha } = kokoro
   const repoDir = join(cacheDir, repo)
   const SENTINELS = {
-    'model.int8.onnx': 'bda15858163726a492d02a9a727bc263551b86ac77f90812c4b30ff41d380e26',
+    [main]: sha,
     'voices.bin': 'e64a5a581d8c2a350d848f51c3121657cd83aa07ed6109172177345874a7244c',
     'tokens.txt': '931ab2df2400cd65d580a22402024c2347ced8ae9ea300e545144b1aacc48e14',
     'lexicon-us-en.txt': '7daaab53a181be9885b853a8582bf1838186317e5dadacbcef9c426d6fa0da14',
     'lexicon-zh.txt': '11111d8cd695fba2ace1367a1d0a708b586e6ef5c1f9be91da5d7eef129b651c',
     'espeak-ng-data/phontab': '886f3fa402cb0ba73d483aa8ad000af47a6b7cc06293c75a97913fba68a530f6',
   }
-  const sentinelOk = await Promise.all(Object.entries(SENTINELS).map(async ([f, sha]) => (await sha256OfFile(join(repoDir, f)).catch(() => '')) === sha))
+  const sentinelOk = await Promise.all(Object.entries(SENTINELS).map(async ([f, sha2]) => (await sha256OfFile(join(repoDir, f)).catch(() => '')) === sha2))
   if (!sentinelOk.every(Boolean)) {
-    console.log(`Kokoro 模型（${repo}）：文件不全，走 HF 树枚举下载（约 400 个文件）`)
+    console.log(`Kokoro 模型（${repo}）：文件不全，走 HF 树枚举下载`)
     const treeRes = await fetch(`${HOST_FALLBACK}/api/models/${repo}/tree/main?recursive=true`, { headers: { 'user-agent': 'dsh-voice-mode/prefetch' } })
     if (treeRes.ok) {
       const tree = await treeRes.json()