dsh-voice-mode 0.7.1 → 0.7.3

package/README.en.md

@@ -11,7 +11,7 @@
 >
 > 中文说明见 [README.md](./README.md)。
 
-> **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.
+> **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 1500 ms.
 
 ## Fork enhancements (this repo)
 
@@ -19,7 +19,7 @@
 - **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;
-- **Long segments**: hold up to 10 min (pauses don't split), continuous sentences up to 3 min, 700 ms silence split by default;
+- **Long segments**: continuous listening stitches consecutive segments into one message (internally chunked at 30 s and concatenated across chunks); 1500 ms silence split by default; hold keeps pauses from splitting;
 - **Hardening**: session-existence check, loopback + Origin guards, per-endpoint rate limits, model SHA256 pinning, download-host allowlist.
 
 > ⚠️ The screenshots below (and `assets/demo.gif`) show the **upstream legacy single-button UI**; the current UI adds a mode-switch button next to the mic.
@@ -29,7 +29,7 @@
 
 - **Voice mode**: toggle with the microphone button in the input toolbar or the global shortcut `Ctrl+Shift+V`; globally single-active (only one session is in voice mode at a time; switching sessions yields automatically)
 - **Two interaction modes (switchable in settings, plus a mode-switch button beside the mic)**:
-  - `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
+  - `toggle` (default) continuous listening: RMS VAD segmentation → streaming zipformer2 ASR (words appear as you speak, live caption preview) → automatic sentence split after 1500 ms of silence into the draft, consecutive segments joined into one message, then auto-sent after ~1500 ms more of silence (≈3 s total); 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/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
@@ -44,7 +44,7 @@
 | Gesture | Behaviour |
 | --- | --- |
 | Click the mic button / `Ctrl+Shift+V` | Enter / exit voice mode |
-| Just speak, pause 700 ms (toggle) | Auto sentence split and send |
+| Just speak, pause ~1500 ms (toggle) | Accumulate into draft, auto-send after ~3 s of quiet |
 | Hold `Ctrl` (toggle, ≥250 ms speech) | Force-send the current segment immediately |
 | **Hold the mic button (hold)** | Hold to talk, release to send; swipe up / `Esc` / blur abandons the segment; <250 ms tap exits the mode |
 | Hold `Ctrl` (hold, ≥600 ms) | Keyboard push-to-talk, release to send |
@@ -86,7 +86,7 @@ npm run prefetch          # run inside the plugin dir; writes to the platform ca
 ## Usage
 
 1. Click the mic button in the input toolbar (or press `Ctrl+Shift+V`) to enter voice mode; a status bar appears above the input box
-2. Choose how to speak: just talk and let the 700 ms pause auto-send (toggle); or hold the mic button and release to send (hold)
+2. Choose how to speak: just talk and let the ~1500 ms pause split and ~3 s of quiet auto-send (toggle); or hold the mic button and release to send (hold)
 3. The AI answer is read sentence-by-sentence with a caption overlay at the bottom-right; click "Skip" or just start speaking to interrupt
 4. Click "Exit" in the status bar (or press `Ctrl+Shift+V` again) to leave voice mode
 
@@ -103,11 +103,11 @@ If a wake word is configured, you land in standby first (the status bar prompts
 | `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 |
-| `silenceMs` | `700` | Silence pause in ms that marks the end of a complete sentence |
+| `silenceMs` | `1500` | Silence pause in ms that marks the end of a complete sentence |
 | `idleTimeoutMinutes` | `10` | Minutes of inactivity before auto-exiting voice mode (reading counts as activity) |
 | `modelHost` | default | Model download host (use `https://hf-mirror.com` on mainland networks) |
 | `autoSend` | `true` | Auto-send after a finalized transcript; when off, text only goes to the draft (hold `Ctrl` / release in hold mode still sends) |
-| `mode` | `toggle` | Interaction mode: `toggle` continuous listening + 700 ms silence split; `hold` push-to-talk, release to send (short tap exits) |
+| `mode` | `toggle` | Interaction mode: `toggle` continuous listening + 1500 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`/`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.
@@ -145,7 +145,7 @@ You can also edit the `voice-mode:` section of `~/.dsh/settings.yaml` directly (
     voice: zh-CN-XiaoxiaoNeural
     rate: 1.0
     interruptLevel: 0
-    silenceMs: 700
+    silenceMs: 1500
     idleTimeoutMinutes: 10
     modelHost: https://huggingface.co
 ```
@@ -181,7 +181,7 @@ You can also edit the `voice-mode:` section of `~/.dsh/settings.yaml` directly (
 ![architecture](https://raw.githubusercontent.com/qishuilalala/dsh-voice-mode/HEAD/plugin/dsh-voice-mode/assets/architecture.svg)
 
 ```
-input:  mic ──RMS VAD (700 ms silence split)──▶ POST /voice-mode/asr (f32 PCM, 16k, incremental)
+input:  mic ──RMS VAD (1500 ms silence split)──▶ POST /voice-mode/asr (f32 PCM, 16k, incremental)
                                             │ zipformer2 streaming ASR (host-side WASM)
                                             ▼
         composer draft ──autoSend──▶ model stream ──llm/stream tap (active voice session only)

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（0.6.0）**：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。
+> **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 1500 毫秒。
 
 ## Fork 增强（本仓库新增）
 
@@ -26,7 +26,7 @@ DeepSeek Harness 语音双工对话模式：会话内一键进入 → 边说边
 - **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）；
 - **交互增强**：输入框旁**模式切换按钮**（持续聆听 ⇄ 按住说话，保存到设置）；按住说模式下**按住才录、不按住不打断**；
-- **长段支持**：按住说整段上限 10 分钟（停顿不断句）；持续聆听单句上限 3 分钟；静音断句默认 700 毫秒；
+- **长段支持**：持续聆听连续多段自动拼成一条消息（内部按 30s 分块识别，跨块累积拼接），静音断句默认 1500 毫秒；按住说停顿不断句；
 - **朗读稳定性**：打断即终止在途合成释放 CPU；句间不再有 3-5 秒停顿；长朗读不触发空闲下线；
 - **安全加固**：会话存在性校验 / 回环+Origin 校验 / 全端点限流 / **ASR+TTS 全模型 SHA256 固定** / 下载域名白名单 / 重定向守卫。
 
@@ -36,7 +36,7 @@ DeepSeek Harness 语音双工对话模式：会话内一键进入 → 边说边
 
 - **语音模式**：输入框工具排麦克风按钮或全局快捷键 `Ctrl+Shift+V` 进入/退出；全局单活（同一时刻仅一个会话处于语音模式，切换会话自动让出）
 - **两种交互模式（输入框旁按钮或设置可切换，切换即持久化）**：
-  - `toggle`（默认）持续聆听：RMS VAD 分段 → zipformer2 流式识别（边说边出字，实时字幕预览）→ 静音约 700 毫秒自动断句进草稿并自动发送；按住 `Ctrl` 强制立即发送
+  - `toggle`（默认）持续聆听：RMS VAD 分段 → zipformer2 流式识别（边说边出字，实时字幕预览）→ 静音约 1500 毫秒断句进草稿，连续多段拼成一条消息，再静音约 1500 毫秒（合计约 3 秒）自动发送；按住 `Ctrl` 强制立即发送
   - `hold` 按住说话：短按进入/退出，**按住麦克风按钮说话、松手即发**（滑出取消、`Esc`/失焦放弃本段）；按住期间停顿不断句（上限 10 分钟）；`Ctrl` 按住即录、松开即发
 - **唤醒词（可选，默认关）**：设置 `wakeWord` 后进入待机态，说出唤醒词才开始识别（如「你好小D」）
 - **输出链路**：只朗读最终答复的 `text-delta`（reasoning/工具调用不读），按句流式朗读（默认 Edge 云端；可切本地 VITS/Kokoro，中英混读选 Kokoro）+ 右下角实时字幕浮层；工具调用触发提示音；全文照常写入聊天记录；口语化提示词（设置 `spokenFormat`，默认开）让回复为自然短句、不带 Markdown 排版符号
@@ -60,7 +60,7 @@ bundle 插件安装后需重启 dsh 生效（Linux：`systemctl restart dsh`；
 | 手势 | 作用 |
 | --- | --- |
 | `Ctrl+Shift+V` | 进入 / 退出语音模式 |
-| 直接说话 | `toggle`：边说边出字，停顿约 700 毫秒自动发送；按住 `Ctrl` 强制立即发送 |
+| 直接说话 | `toggle`：边说边出字，停顿约 1500 毫秒断句进草稿、再静音约 1500 毫秒（合计约 3 秒）自动发送；按住 `Ctrl` 强制立即发送 |
 | 按住麦克风按钮 | `hold`：松手发送；短按退出；滑出 / `Esc` / 失焦放弃本段 |
 | 点输入框旁模式按钮 | 在「持续聆听 ⇄ 按住说」间切换（保存到设置） |
 | 说唤醒词 | 待机态激活识别（配置后） |
@@ -77,11 +77,11 @@ bundle 插件安装后需重启 dsh 生效（Linux：`systemctl restart dsh`；
 | `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 |
-| `silenceMs` | `700` | 说完整一句的静音停顿毫秒数 |
+| `silenceMs` | `1500` | 说完整一句的静音停顿毫秒数 |
 | `idleTimeoutMinutes` | `10` | 无活动自动退出语音模式的分钟数（朗读计为活动） |
 | `modelHost` | 默认源 | 模型下载源（国内网络填 `https://hf-mirror.com`） |
 | `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） |
+| `mode` | `toggle` | 交互模式：`toggle` 持续聆听 + 1500ms 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` | `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** |
 

package/lib/client.js

@@ -992,7 +992,6 @@ function createAsrEngine(config, sessionId) {
     beginHeld() {
       if (!active || holdActive) return;
       holdActive = true;
-      forcePending = true;
       segmentEpoch++;
       utteranceEndAt = null;
       segment = [];
@@ -1307,10 +1306,10 @@ var zh = {
   sev0: "0 \u9AD8\u95E8\u69DB",
   sev1: "1 \u4E2D",
   sev2: "2 \u4F4E",
-  descSilence: "\u8BF4\u5B8C\u6574\u4E00\u53E5\u7684\u9759\u97F3\u505C\u987F\u6BEB\u79D2\u6570\uFF08\u9ED8\u8BA4 700 \u6BEB\u79D2\uFF1B\u81F3\u5C11 250ms \u8BED\u97F3\u624D\u5224\u53E5\uFF0C\u9632\u77ED\u4FC3\u566A\u58F0\u8BEF\u89E6\u53D1\uFF09",
+  descSilence: "\u8BF4\u5B8C\u6574\u4E00\u53E5\u7684\u9759\u97F3\u505C\u987F\u6BEB\u79D2\u6570\uFF08\u9ED8\u8BA4 1500 \u6BEB\u79D2\uFF0C\u7ED9\u601D\u8003\u505C\u987F\u7559\u7A7A\u95F4\uFF1B\u81F3\u5C11 250ms \u8BED\u97F3\u624D\u5224\u53E5\uFF0C\u9632\u77ED\u4FC3\u566A\u58F0\u8BEF\u89E6\u53D1\uFF09",
   descIdle: "\u65E0\u6D3B\u52A8\u81EA\u52A8\u9000\u51FA\u8BED\u97F3\u6A21\u5F0F\u7684\u5206\u949F\u6570\uFF08\u9ED8\u8BA4 10\uFF09",
   descModelHost: "ASR \u6A21\u578B\u4E0B\u8F7D\u6E90\uFF08\u5B98\u65B9\u6E90 / \u56FD\u5185\u955C\u50CF\uFF0C\u6216\u9009\u300C\u81EA\u5B9A\u4E49\u300D\u586B\u4EFB\u610F\u955C\u50CF\uFF09",
-  descAutoSend: "\u8BC6\u522B\u5B9A\u7A3F\u540E\u81EA\u52A8\u53D1\u9001\uFF08\u5173=\u53EA\u8FDB\u8349\u7A3F\uFF1B\u6309\u4F4F Ctrl / hold \u677E\u624B\u4ECD\u53D1\u9001\uFF09",
+  descAutoSend: "\u9759\u97F3\u5230\u70B9\u81EA\u52A8\u53D1\u9001\uFF08\u8FDE\u7EED\u591A\u6BB5\u62FC\u6210\u4E00\u6761\uFF1B\u5173=\u53EA\u8FDB\u8349\u7A3F\uFF1B\u6309\u4F4F Ctrl / hold \u677E\u624B\u4ECD\u53D1\u9001\uFF09",
   descAutoResume: "\u5207\u6362\u56DE\u4E0A\u6B21\u8BED\u97F3\u4F1A\u8BDD\u65F6\u81EA\u52A8\u6062\u590D\u8BED\u97F3\u6A21\u5F0F\uFF08\u9ED8\u8BA4\u5173\uFF0C\u9700\u9EA6\u514B\u98CE\u6743\u9650\u5DF2\u6388\u4E88\uFF1B\u7701\u53BB\u6BCF\u6B21\u5207\u6362\u4F1A\u8BDD\u540E\u91CD\u65B0\u70B9\u9EA6\u514B\u98CE\uFF09",
   descSpokenFormat: "\u8BED\u97F3\u4F1A\u8BDD\u6CE8\u5165\u53E3\u8BED\u5316\u63D0\u793A\u8BCD\uFF08\u56DE\u590D\u53E3\u8BED\u5316\u3001\u77ED\u53E5\u3001\u4E0D\u7528 Markdown \u6392\u7248\u7B26\u53F7\uFF0C\u6717\u8BFB\u66F4\u987A\u66F4\u5FEB\uFF1B\u9ED8\u8BA4\u5F00\uFF0C\u6539\u52A8\u5373\u65F6\u751F\u6548\uFF09",
   descSenseVoice: "\u5B9A\u7A3F\u7528 SenseVoice \u91CD\u8BD1\uFF08\u5E26\u6807\u70B9 + \u6570\u5B57\u5F52\u4E00\u5316\u3001\u8BC6\u522B\u66F4\u51C6\uFF1B\u9ED8\u8BA4\u5F00\u3002\u5173\u95ED\u53EF\u7701 228MB \u6A21\u578B\uFF0C\u53EA\u8D70\u6D41\u5F0F\u8BC6\u522B\uFF09",
@@ -1448,10 +1447,10 @@ var en = {
   sev0: "0 high",
   sev1: "1 medium",
   sev2: "2 low",
-  descSilence: "Silence pause before a sentence is committed (default 700 ms; at least 250 ms of speech required, guards against noise triggers)",
+  descSilence: "Silence pause before a sentence is committed (default 1500 ms; at least 250 ms of speech required, guards against noise triggers)",
   descIdle: "Auto-exit voice mode after idle minutes (default 10)",
   descModelHost: "ASR model download source (official source / mirror, or any custom URL)",
-  descAutoSend: "Auto-send after finalized recognition (off = draft only; Ctrl / hold still sends)",
+  descAutoSend: "Auto-send once quiet (consecutive segments join into one message; off = draft only; Ctrl / hold still sends)",
   descAutoResume: "Auto-resume voice mode when switching back to the last voice session (default off, requires granted mic permission)",
   descSpokenFormat: "Inject spoken-format prompt into voice replies (colloquial, short sentences, no Markdown; default on, live)",
   descSenseVoice: "Re-transcribe the finalized utterance with SenseVoice (punctuation + ITN, more accurate; default on \u2014 turn off to skip the 228 MB model and keep streaming only)",
@@ -2458,7 +2457,7 @@ function VoiceSettingsCard({ scope }) {
       /* @__PURE__ */ (0, import_jsx_runtime.jsxs)(Section, { title: t("secRecognition"), children: [
         /* @__PURE__ */ (0, import_jsx_runtime.jsx)(Row, { name: "senseVoice", desc: t("descSenseVoice"), children: /* @__PURE__ */ (0, import_jsx_runtime.jsx)("input", { type: "checkbox", checked: Boolean(value.senseVoice), onChange: (e) => void scope.set("senseVoice", e.target.checked) }) }),
         /* @__PURE__ */ (0, import_jsx_runtime.jsx)(Row, { name: "spokenFormat", desc: t("descSpokenFormat"), children: /* @__PURE__ */ (0, import_jsx_runtime.jsx)("input", { type: "checkbox", checked: Boolean(value.spokenFormat), onChange: (e) => void scope.set("spokenFormat", e.target.checked) }) }),
-        /* @__PURE__ */ (0, import_jsx_runtime.jsx)(Row, { name: "silenceMs", desc: t("descSilence"), children: /* @__PURE__ */ (0, import_jsx_runtime.jsx)(NumberField, { score: scope, field: "silenceMs", value: value.silenceMs ?? 700, min: 500, max: 3e4, step: 100 }) }),
+        /* @__PURE__ */ (0, import_jsx_runtime.jsx)(Row, { name: "silenceMs", desc: t("descSilence"), children: /* @__PURE__ */ (0, import_jsx_runtime.jsx)(NumberField, { score: scope, field: "silenceMs", value: value.silenceMs ?? 1500, min: 500, max: 3e4, step: 100 }) }),
         /* @__PURE__ */ (0, import_jsx_runtime.jsx)(Row, { name: "idleTimeoutMinutes", desc: t("descIdle"), children: /* @__PURE__ */ (0, import_jsx_runtime.jsx)(NumberField, { score: scope, field: "idleTimeoutMinutes", value: value.idleTimeoutMinutes ?? 10, min: 1, max: 120, step: 1 }) })
       ] }),
       /* @__PURE__ */ (0, import_jsx_runtime.jsx)(Section, { title: t("secModel"), children: /* @__PURE__ */ (0, import_jsx_runtime.jsx)(Row, { name: "modelHost", desc: t("descModelHost"), children: /* @__PURE__ */ (0, import_jsx_runtime.jsx)(SelectField, { score: scope, field: "modelHost", value: value.modelHost ?? "", options: HOST_OPTIONS, placeholder: "https://..." }) }) }),
@@ -2485,7 +2484,7 @@ var TELEMETRY_VIEW = [
   { stage: "first-tts-chunk", key: "telFirstChunk" },
   { stage: "first-audio-played", key: "telFirstPlayed" }
 ];
-var BUILD_TAG = "697c538";
+var BUILD_TAG = "4ab9c0d";
 var TELEMETRY_FLAG = "dsh-voice-mode.telemetry";
 var telemetryEnabled = typeof localStorage !== "undefined" && localStorage.getItem(TELEMETRY_FLAG) === "1";
 console.log("[dsh-voice] build=" + BUILD_TAG);
@@ -2778,7 +2777,7 @@ function createVoiceBus(basePath = BASE_PATH2, ctx) {
   let activeSessionId = null;
   const DEFAULT_BOOT = {
     basePath: BASE_PATH2,
-    silenceMs: 700,
+    silenceMs: 1500,
     interruptLevel: 0,
     idleTimeoutMinutes: 10,
     autoSend: true,
@@ -3241,6 +3240,7 @@ function MicButton({
   const engineRef = (0, import_react2.useRef)(null);
   const actionsRef = (0, import_react2.useRef)(inputActions);
   const submitTimerRef = (0, import_react2.useRef)(null);
+  const autoSendTimerRef = (0, import_react2.useRef)(null);
   const idleTimerRef = (0, import_react2.useRef)(null);
   const runningRef = (0, import_react2.useRef)(false);
   const mountedRef = (0, import_react2.useRef)(true);
@@ -3248,7 +3248,7 @@ function MicButton({
   const manualHoldRef = (0, import_react2.useRef)(false);
   const breakRef = (0, import_react2.useRef)(null);
   const pausedForHiddenRef = (0, import_react2.useRef)(false);
-  const bootNow = () => bus.ui.boot ?? { basePath: "/voice-mode", silenceMs: 700, interruptLevel: 0, idleTimeoutMinutes: 10, autoSend: true, autoResume: false, mode: "toggle", bargeInMode: "auto", echoGateDb: 6, shortcut: "Ctrl+Shift+V", wakeWord: "", toolBeep: false };
+  const bootNow = () => bus.ui.boot ?? { basePath: "/voice-mode", silenceMs: 1500, interruptLevel: 0, idleTimeoutMinutes: 10, autoSend: true, autoResume: false, mode: "toggle", bargeInMode: "auto", echoGateDb: 6, shortcut: "Ctrl+Shift+V", wakeWord: "", toolBeep: false };
   useVoiceCss();
   const [, bumpUi] = (0, import_react2.useState)(0);
   (0, import_react2.useEffect)(
@@ -3309,6 +3309,7 @@ function MicButton({
         setLocalMode("off");
         clearIdle();
         cancelPendingSubmit();
+        cancelAutoSend();
         clearBreakTimer();
         setHolding(false);
         isSpeechTrueCount = 0;
@@ -3329,11 +3330,61 @@ function MicButton({
       submitTimerRef.current = null;
     }
   };
+  const cancelAutoSend = () => {
+    if (autoSendTimerRef.current) {
+      clearTimeout(autoSendTimerRef.current);
+      autoSendTimerRef.current = null;
+    }
+  };
+  const submitDraftNow = (expectedText) => {
+    cancelPendingSubmit();
+    const actions = actionsRef.current;
+    const submitFn = actions?.submit;
+    if (typeof submitFn !== "function") return;
+    const draftSnapshot = (expectedText ?? draftRef.current).trim();
+    if (!draftSnapshot) return;
+    const doSubmit = () => {
+      try {
+        const r = submitFn();
+        if (r && typeof r.then === "function") {
+          r.catch(() => {
+            bus.setUi({ error: t("sendFailKept") });
+          });
+        }
+      } catch {
+        bus.setUi({ error: t("sendFailKept") });
+      }
+    };
+    doSubmit();
+    let retryCount = 0;
+    submitTimerRef.current = setInterval(() => {
+      retryCount++;
+      const phase = phaseRef.current;
+      if (retryCount > 3 || phase === "submitting" || phase === "adjudicating" || draftRef.current.trim() !== draftSnapshot) {
+        cancelPendingSubmit();
+        return;
+      }
+      doSubmit();
+    }, 500);
+  };
+  const scheduleAutoSend = () => {
+    cancelAutoSend();
+    const delay = bootNow().silenceMs;
+    autoSendTimerRef.current = setTimeout(() => {
+      autoSendTimerRef.current = null;
+      const eng = engineRef.current;
+      if (eng && (eng.state === "speech" || eng.holding)) return;
+      if (bus.ui.playing) return;
+      if (bootNow().autoSend === false) return;
+      submitDraftNow();
+    }, delay);
+  };
   const exitMode = async (_reason) => {
     if (localRef.current === "off") return;
     setLocalMode("off");
     clearIdle();
     cancelPendingSubmit();
+    cancelAutoSend();
     isSpeechTrueCount = 0;
     fixtureRecorder.save("exit");
     breakRef.current = null;
@@ -3521,6 +3572,7 @@ function MicButton({
       }
       engine.onState((s) => {
         bus.setUi({ state: s });
+        if (s === "speech") cancelAutoSend();
         if (s === "idle") resetIdle();
       });
       engine.onError((key) => {
@@ -3538,45 +3590,27 @@ function MicButton({
         const actions = actionsRef.current;
         const trimmed = text.trim();
         if (!trimmed) return;
+        let nextDraft = trimmed;
         try {
           const curText = draftRef.current;
-          const nextDraft = curText ? `${curText} ${trimmed}` : trimmed;
+          nextDraft = curText ? `${curText} ${trimmed}` : trimmed;
           if (typeof actions?.setDraft === "function") actions.setDraft(nextDraft);
           else if (typeof actions?.setDraft === "function") actions.setDraft(nextDraft);
-          else {
-          }
         } catch {
           try {
             actions?.setDraft?.(trimmed);
           } catch {
           }
         }
-        if (bus.ui.playing && !meta?.force) return;
-        if (bootNow().autoSend === false && !meta?.force) return;
-        const doSubmit = () => {
-          try {
-            const r = actions?.submit?.();
-            if (r && typeof r.then === "function") {
-              r.catch(() => {
-                bus.setUi({ error: t("sendFailKept") });
-              });
-            }
-          } catch {
-            bus.setUi({ error: t("sendFailKept") });
-          }
-        };
-        cancelPendingSubmit();
-        doSubmit();
-        let retryCount = 0;
-        submitTimerRef.current = setInterval(() => {
-          retryCount++;
-          const phase = phaseRef.current;
-          if (retryCount > 3 || phase === "submitting" || phase === "adjudicating" || draftRef.current.trim() !== trimmed) {
-            cancelPendingSubmit();
-            return;
-          }
-          doSubmit();
-        }, 500);
+        if (meta?.force) {
+          cancelAutoSend();
+          submitDraftNow(nextDraft);
+          return;
+        }
+        if (bootNow().mode === "hold") return;
+        if (bus.ui.playing) return;
+        if (bootNow().autoSend === false) return;
+        scheduleAutoSend();
       });
       bus.setUi({ state: "idle", partial: "", levels: [], error: null, model: null, ttsNotice: null });
       if (!mountedRef.current) {
@@ -3650,6 +3684,7 @@ function MicButton({
       mountedRef.current = false;
       clearIdle();
       cancelPendingSubmit();
+      cancelAutoSend();
       isSpeechTrueCount = 0;
       const sid = sidRef.current;
       if ((localRef.current === "on" || localRef.current === "pending") && sid) {
@@ -4040,7 +4075,7 @@ function VoiceStatusBar({ bus, sessionId }) {
   if (b.ui.interruptConfirmMs !== void 0) {
     telParts.push(`${t("interruptConfirm")} ${fmt(b.ui.interruptConfirmMs)}`);
   }
-  if (b.ui.echoLevels) {
+  if (telemetryEnabled && b.ui.echoLevels) {
     const el = b.ui.echoLevels;
     telParts.push(
       `AEC delay=${Math.round(b.ui.echoDelayMs ?? 0)}ms floor=${el.floorRms.toFixed(4)} resid=${el.residualRms.toFixed(4)}`

package/lib/index.js

@@ -392,7 +392,7 @@ function rmsOf(samples) {
   return Math.sqrt(sum / samples.length);
 }
 function createAsrRuntime(options) {
-  const { cacheDir, modelHost, broadcast, senseVoice, allowCustomHost } = options;
+  const { cacheDir, modelHost, broadcast, senseVoice, silenceMs, allowCustomHost } = options;
   let lastProgress = null;
   const localBroadcast = (event, payload) => {
     if (event === "asr-progress") lastProgress = payload;
@@ -473,11 +473,11 @@ function createAsrRuntime(options) {
     }
     return vadLoading;
   };
-  const newVad = (vadPath, threshold = 0.5) => createVad({
+  const newVad = (vadPath, threshold = 0.5, minSilenceDuration = 0.5) => createVad({
     sileroVad: {
       model: vadPath,
       threshold,
-      minSilenceDuration: 0.5,
+      minSilenceDuration,
       minSpeechDuration: 0.25,
       maxSpeechDuration: 20,
       windowSize: 512
@@ -492,7 +492,7 @@ function createAsrRuntime(options) {
     if (seg.vad) return seg.vad;
     const vadPath = await ensureVadModel();
     if (!vadPath) return null;
-    seg.vad = newVad(vadPath);
+    seg.vad = newVad(vadPath, 0.5, silenceMs() / 1e3);
     return seg.vad;
   };
   const detectVads = /* @__PURE__ */ new Map();
@@ -1184,44 +1184,54 @@ var TtsQueue = class {
     try {
       while (q.pending.length > 0) {
         const item = q.pending.shift();
-        try {
-          const buf = await this.engine.synthesize(item.text);
-          if (item.epoch !== q.epoch) continue;
-          q.errorNotified = false;
-          q.backoff = 0;
-          const sentenceId = q.seq++;
-          const mime = this.engine.mime;
-          const dataFrame = {
-            sessionId,
-            sentenceId,
-            chunkId: 0,
-            final: false,
-            audio: buf.toString("base64"),
-            mime
-          };
-          for (const fn of this.listeners) {
-            try {
-              fn(dataFrame);
-            } catch {
+        const MAX_SYNTH_ATTEMPTS = 3;
+        let buf = null;
+        for (let attempt = 0; attempt < MAX_SYNTH_ATTEMPTS; attempt++) {
+          if (item.epoch !== q.epoch) break;
+          try {
+            buf = await this.engine.synthesize(item.text);
+            break;
+          } catch (e) {
+            console.warn(`[dsh-voice-mode] synthesis failed (${attempt + 1}/${MAX_SYNTH_ATTEMPTS}): ${String(e)}`);
+            if (attempt < MAX_SYNTH_ATTEMPTS - 1) {
+              await new Promise((r) => setTimeout(r, 400 * (attempt + 1)));
             }
           }
-          const finalFrame = {
-            sessionId,
-            sentenceId,
-            chunkId: 1,
-            final: true,
-            text: item.text,
-            audio: "",
-            mime
-          };
-          for (const fn of this.listeners) {
-            try {
-              fn(finalFrame);
-            } catch {
-            }
+        }
+        if (item.epoch !== q.epoch) continue;
+        if (buf === null) continue;
+        q.errorNotified = false;
+        q.backoff = 0;
+        const sentenceId = q.seq++;
+        const mime = this.engine.mime;
+        const dataFrame = {
+          sessionId,
+          sentenceId,
+          chunkId: 0,
+          final: false,
+          audio: buf.toString("base64"),
+          mime
+        };
+        for (const fn of this.listeners) {
+          try {
+            fn(dataFrame);
+          } catch {
+          }
+        }
+        const finalFrame = {
+          sessionId,
+          sentenceId,
+          chunkId: 1,
+          final: true,
+          text: item.text,
+          audio: "",
+          mime
+        };
+        for (const fn of this.listeners) {
+          try {
+            fn(finalFrame);
+          } catch {
           }
-        } catch (e) {
-          console.warn(`[dsh-voice-mode] synthesis failed: ${String(e)}`);
         }
       }
     } catch (e) {
@@ -1773,7 +1783,7 @@ var VOICE_SETTINGS_DEFAULTS = {
   voice: "zh-CN-XiaoxiaoNeural",
   rate: 1,
   interruptLevel: 0,
-  silenceMs: 700,
+  silenceMs: 1500,
   idleTimeoutMinutes: 10,
   modelHost: "",
   autoSend: true,
@@ -1801,10 +1811,10 @@ function createVoiceSettingsSchema(defs) {
     ),
     rate: z.number().min(0.5).max(2).default(d.rate).description("\u6717\u8BFB\u8BED\u901F\u500D\u7387\uFF080.5 = \u6162\u901F\uFF0C2.0 = \u5FEB\u901F\uFF0C1.0 = \u6B63\u5E38\uFF09"),
     interruptLevel: z.union([z.const(0), z.const(1), z.const(2)]).default(d.interruptLevel).description("\u53D1\u58F0\u6253\u65AD\u7075\u654F\u5EA6\uFF1A0 \u9AD8\u95E8\u69DB\uFF08\u5B89\u9759\u73AF\u5883\uFF0C\u9ED8\u8BA4\uFF09/ 1 \u4E2D / 2 \u4F4E\uFF08\u5608\u6742\u73AF\u5883\u66F4\u5BB9\u6613\u6253\u65AD\uFF09"),
-    silenceMs: z.number().min(500).max(3e4).default(d.silenceMs).description("\u8BF4\u5B8C\u6574\u4E00\u53E5\u7684\u9759\u97F3\u505C\u987F\u6BEB\u79D2\u6570\uFF08\u9ED8\u8BA4 700 \u6BEB\u79D2\uFF1B\u81F3\u5C11 250ms \u8BED\u97F3\u624D\u5224\u53E5\uFF0C\u9632\u77ED\u4FC3\u566A\u58F0\u8BEF\u89E6\u53D1\uFF09"),
+    silenceMs: z.number().min(500).max(3e4).default(d.silenceMs).description("\u8BF4\u5B8C\u6574\u4E00\u53E5\u7684\u9759\u97F3\u505C\u987F\u6BEB\u79D2\u6570\uFF08\u9ED8\u8BA4 1500 \u6BEB\u79D2\uFF0C\u7ED9\u601D\u8003\u505C\u987F\u7559\u7A7A\u95F4\uFF1B\u81F3\u5C11 250ms \u8BED\u97F3\u624D\u5224\u53E5\uFF0C\u9632\u77ED\u4FC3\u566A\u58F0\u8BEF\u89E6\u53D1\uFF09"),
     idleTimeoutMinutes: z.number().min(1).max(120).default(d.idleTimeoutMinutes).description("\u65E0\u6D3B\u52A8\u81EA\u52A8\u9000\u51FA\u8BED\u97F3\u6A21\u5F0F\u7684\u5206\u949F\u6570\uFF08\u9ED8\u8BA4 10\uFF09"),
     modelHost: z.string().default(d.modelHost).description("ASR \u6A21\u578B\u4E0B\u8F7D\u6E90\uFF08\u7559\u7A7A\u7528\u9ED8\u8BA4\u6E90\uFF1B\u56FD\u5185\u7F51\u7EDC\u53EF\u586B https://hf-mirror.com\uFF09"),
-    autoSend: z.boolean().default(d.autoSend).description("\u8BC6\u522B\u5B9A\u7A3F\u540E\u81EA\u52A8\u53D1\u9001\uFF08\u5173\u95ED\u5219\u53EA\u8FDB\u8349\u7A3F\u4F9B\u7F16\u8F91\uFF1B\u6309\u4F4F Ctrl / hold \u677E\u624B\u4ECD\u4F1A\u53D1\u9001\uFF09"),
+    autoSend: z.boolean().default(d.autoSend).description("\u9759\u97F3\u5230\u70B9\u81EA\u52A8\u53D1\u9001\uFF08\u8FDE\u7EED\u591A\u6BB5\u62FC\u6210\u4E00\u6761\u6D88\u606F\uFF1B\u5173\u95ED\u5219\u53EA\u8FDB\u8349\u7A3F\u4F9B\u7F16\u8F91\uFF1B\u6309\u4F4F Ctrl / hold \u677E\u624B\u4ECD\u4F1A\u53D1\u9001\uFF09"),
     autoResume: z.boolean().default(d.autoResume).description("\u5207\u6362\u56DE\u4E0A\u6B21\u8BED\u97F3\u4F1A\u8BDD\u65F6\u81EA\u52A8\u6062\u590D\u8BED\u97F3\u6A21\u5F0F\uFF08\u9ED8\u8BA4\u5173\uFF0C\u9700\u9EA6\u514B\u98CE\u6743\u9650\u5DF2\u6388\u4E88\uFF1B\u5173\u95ED\u5219\u6BCF\u6B21\u5207\u6362\u4F1A\u8BDD\u540E\u9700\u91CD\u65B0\u70B9\u9EA6\u514B\u98CE\uFF09"),
     mode: z.union([z.const("toggle"), z.const("hold")]).default(d.mode).description("\u4EA4\u4E92\u6A21\u5F0F\uFF1Atoggle \u6301\u7EED\u8046\u542C + \u9759\u97F3\u81EA\u52A8\u65AD\u53E5\uFF08\u9ED8\u8BA4\uFF09\uFF1Bhold \u6309\u4F4F\u8BF4\u8BDD\u3001\u677E\u624B\u53D1\u9001\uFF08\u77ED\u6309\u9000\u51FA\uFF09"),
     bargeInMode: z.union([z.const("auto"), z.const("manual")]).default(d.bargeInMode).description("\u6253\u65AD\u65B9\u5F0F\uFF1Aauto \u81EA\u52A8\u6253\u65AD\uFF08\u5F00\u53E3\u5373\u6253\u65AD\uFF0C\u8033\u673A/\u5B89\u9759\u73AF\u5883\u63A8\u8350\uFF09\uFF1Bmanual \u624B\u52A8\u6253\u65AD\uFF08\u5916\u653E\u63A8\u8350\u2014\u2014\u5916\u653E\u56DE\u58F0\u4F1A\u8BEF\u89E6\u53D1\u81EA\u52A8\u6253\u65AD\uFF0C\u6539\u6309\u4F4F\u9EA6\u514B\u98CE/Ctrl \u663E\u5F0F\u6253\u65AD\uFF0C\u6C38\u4E0D\u81EA\u6253\u65AD\uFF09"),
@@ -1828,7 +1838,7 @@ var Config = z.object({
   voice: z.string().default("zh-CN-XiaoxiaoNeural"),
   rate: z.number().default(1),
   interruptLevel: z.union([z.const(0), z.const(1), z.const(2)]).default(0),
-  silenceMs: z.number().default(700),
+  silenceMs: z.number().default(1500),
   idleTimeoutMinutes: z.number().default(10)
 });
 function apply(ctx, config) {
@@ -1896,6 +1906,8 @@ function apply(ctx, config) {
     modelHost: () => vset.modelHost,
     // P4：SenseVoice 定稿重译开关（实时读取，关闭则不下载/不创建模型）。
     senseVoice: () => vset.senseVoice,
+    // 断句静音阈值（实时读取）：端点 VAD minSilenceDuration 跟随设置。
+    silenceMs: () => vset.silenceMs,
     allowCustomHost: config.allowCustomModelHost,
     broadcast
   });

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "dsh-voice-mode",
   "description": "Full-duplex voice plugin for DeepSeek Harness: local zipformer2 streaming ASR (no API key) → editable draft; Edge TTS or local VITS / Kokoro read-aloud with live captions; true barge-in; hardened HTTP surface + model SHA256 pinning; compatible with dsh 0.1.1 → 0.1.2. · DSH 语音双工插件：流式识别入草稿、按句朗读+实时字幕、开口即打断；免 API Key、安全加固、新旧版本兼容。",
-  "version": "0.7.1",
+  "version": "0.7.3",
   "type": "module",
   "main": "lib/index.js",
   "repository": {