唤醒词文档里 tokens.txt 的行数改成 227

实际包里的 tokens.txt 是 227 行，之前写的 228 行不对，按实测值改过来。
唤醒词文档改准：int8 的定位和 decoder 的版本
2026-10-27 10:23:37 +08:00 · 2026-10-27 09:13:42 +08:00 · 2026-10-26 22:12:21 +08:00 · 2026-10-26 21:38:47 +08:00 · 2026-10-26 21:32:37 +08:00 · 2026-10-26 21:16:47 +08:00
13 changed files with 160 additions and 142 deletions
@@ -212,7 +212,7 @@ Websocket接口地址: wss://2662r3426b.vicp.fun/xiaozhi/v1/

 | 工具名称 | 位置 | 使用方法 | 功能说明 |
 |:---:|:---|:---:|:---:|
-| 音频交互测试工具 | main》digital-human》index.html | 在 `main/digital-human` 按其 README 装好依赖后执行 `python start.py`，然后访问 `http://127.0.0.1:8006/index.html` | 测试音频播放和接收功能，验证Python端音频处理是否正常 |
+| 音频交互测试工具 | main》digital-human》index.html | 在 `main/digital-human` 执行 `python start.py` 后访问 `http://127.0.0.1:8006/index.html` | 测试音频播放和接收功能，验证Python端音频处理是否正常 |
 | 模型响应测试工具 | main》xiaozhi-server》performance_tester.py | 执行 `python performance_tester.py` | 测试ASR(语音识别)、LLM(大模型)、VLLM(视觉模型)、TTS(语音合成)等核心模块的响应速度 |

 > 💡 提示：测试模型速度时，只会测试配置了密钥的模型。
@@ -256,7 +256,7 @@ Websocket接口地址: wss://2662r3426b.vicp.fun/xiaozhi/v1/

 | 使用方式 | 支持平台 | 免费平台 |
 |:---:|:---:|:---:|
-| openai 接口调用 | 阿里百炼、火山引擎、DeepSeek、智谱、Gemini、科大讯飞、LM Studio | 智谱、Gemini |
+| openai 接口调用 | 阿里百炼、火山引擎、DeepSeek、智谱、Gemini、科大讯飞 | 智谱、Gemini |
 | ollama 接口调用 | Ollama | - |
 | dify 接口调用 | Dify | - |
 | fastgpt 接口调用 | Fastgpt | - |
@@ -300,7 +300,7 @@ Websocket接口地址: wss://2662r3426b.vicp.fun/xiaozhi/v1/
 | 使用方式 | 支持平台 | 免费平台 |
 |:---:|:---:|:---:|
 | 本地使用 | FunASR、SherpaASR、SherpaParaformerASR、VoskASR | FunASR、SherpaASR、SherpaParaformerASR、VoskASR |
-| 接口调用 | FunASRServer、火山引擎、科大讯飞、腾讯云、阿里云及百炼、百度云、OpenAI ASR、GroqASR、Qwen3ASRFlash | FunASRServer |
+| 接口调用 | FunASRServer、火山引擎、科大讯飞、腾讯云、阿里云、百度云、OpenAI ASR、GroqASR、Qwen3ASRFlash | FunASRServer |

 ---

@@ -28,16 +28,22 @@ wget https://github.com/k2-fsa/sherpa-onnx/releases/download/kws-models/sherpa-o
 tar xvf sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01.tar.bz2

 # 方法2：使用ModelScope
+# 这份仓库里只有 epoch-12 的模型，文件名里的 avg 是 avg-2，不是 avg-1
+# 照抄下面「配置模型文件」的命令时，把 epoch-99-avg-1 换成 epoch-12-avg-2（avg 一起换，只换 epoch 数字会找不到文件）
+# int8 只决定你选方案一还是方案二，不是方法2 的前提：这份仓库里 epoch-12 的 fp32 三件套齐全，
+# 用方法2 + 方案一（精度优先）照样能配出四个必需文件
 pip install modelscope
 python -c "
 from modelscope import snapshot_download
-snapshot_download('pkufool/sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01', cache_dir='./models')
+snapshot_download('pkufool/sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01', local_dir='./sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01')
 "
 ```

+运行时只认 `models/` 目录下的四个固定文件名（`encoder.onnx`、`decoder.onnx`、`joiner.onnx`、`tokens.txt`），也不搜子目录（判定见 `wakeword_runtime/core/detector_assets.py`）。文件留在下载目录里不做拷贝改名，启动会报 `No valid model directory found`。两种方法下载完都要接着做下面「配置模型文件」那节的拷贝改名。
+
 #### 2. 配置模型文件

-模型包下载后包含以下文件：
+模型包下载后包含以下文件（epoch-99 那几行只有官方包有，ModelScope 那份只有 epoch-12）：

 ```
 sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01/
@@ -45,7 +51,9 @@ sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01/
 ├── encoder-epoch-12-avg-2-chunk-16-left-64.onnx
 ├── encoder-epoch-99-avg-1-chunk-16-left-64.int8.onnx    # 速度优先
 ├── encoder-epoch-99-avg-1-chunk-16-left-64.onnx         # 精度优先
+├── decoder-epoch-12-avg-2-chunk-16-left-64.int8.onnx    # 速度优先
 ├── decoder-epoch-12-avg-2-chunk-16-left-64.onnx
+├── decoder-epoch-99-avg-1-chunk-16-left-64.int8.onnx    # 速度优先
 ├── decoder-epoch-99-avg-1-chunk-16-left-64.onnx         # 精度优先
 ├── joiner-epoch-12-avg-2-chunk-16-left-64.int8.onnx     # 速度优先
 ├── joiner-epoch-12-avg-2-chunk-16-left-64.onnx
@@ -87,7 +95,8 @@ cd sherpa-onnx-kws-zipformer-wenetspeech-3.3M-2024-01-01
 # 创建模型目录
 mkdir -p ../models

-# 复制速度优先的epoch-99 int8三件套
+# 复制速度优先的epoch-99 三件套：encoder 和 joiner 用 int8，decoder 用 fp32
+# 官方包里 decoder 也有 int8 版，但量化对 decoder 收益不大，一般仍用 fp32
 cp encoder-epoch-99-avg-1-chunk-16-left-64.int8.onnx ../models/encoder.onnx
 cp decoder-epoch-99-avg-1-chunk-16-left-64.onnx ../models/decoder.onnx
 cp joiner-epoch-99-avg-1-chunk-16-left-64.int8.onnx ../models/joiner.onnx
@@ -98,9 +107,10 @@ cp tokens.txt ../models/tokens.txt

 **注意事项**:

- **不要混用 fp32 与 int8**：三个模型文件必须保持一致的精度
+- **精度怎么选**：encoder、joiner、decoder 三件套都有 int8 和 fp32 两版，官方包和 ModelScope 那份都是这样；官方文档的说法是量化对 decoder 收益不大，所以方案二里 encoder 和 joiner 用 int8、decoder 仍用 fp32
+- **int8 只决定选方案一还是方案二**：运行时只按上面的文件名找模型，不看你拷的是 int8 还是 fp32，所以方法2（ModelScope）配方案一（精度优先）照样能跑
 - **优先选择 epoch-99**：比 epoch-12 训练更充分，精度更高
- **必需文件**：`encoder.onnx` + `decoder.onnx` + `joiner.onnx` + `tokens.txt` + `keywords.txt`
+- **必需文件**：`encoder.onnx` + `decoder.onnx` + `joiner.onnx` + `tokens.txt` 四个，`keywords.txt` 不在其中，它不用自己准备，首次启动会用配置里的唤醒词自动生成

 ### 最终模型文件结构

@@ -111,7 +121,7 @@ wakeword_runtime/models/
 ├── encoder.onnx      # 编码器模型（重命名后）
 ├── decoder.onnx      # 解码器模型（重命名后）
 ├── joiner.onnx       # 连接器模型（重命名后）
-├── tokens.txt        # 拼音 Token 映射表（228行版本）
+├── tokens.txt        # 拼音 Token 映射表（227行版本）
 ├── keywords.txt      # 关键词配置文件（首次启动自动生成）
 └── keywords_raw.txt  # 可选，runtime 不依赖
 ```
@@ -194,12 +204,14 @@ python start.py

 ### 首次使用

-1. 准备 `models/` 目录下的模型文件和 `tokens.txt`
-2. 确认 `models/keywords.txt` 存在
+1. 先按上面「模型下载（必需）」那节把模型包下载解压，再按「配置模型文件」那节把模型放到 `models/` 目录，该目录下要有 `encoder.onnx`、`decoder.onnx`、`joiner.onnx` 和 `tokens.txt` 四个文件
+2. `models/keywords.txt` 不用自己准备，首次启动会用配置里的唤醒词自动生成
 3. 在 `digital-human` 目录运行 `python start.py`
 4. 浏览器打开 `http://127.0.0.1:8006/index.html`
 5. 进入设置页检查“唤醒词”配置

+模型没放到位就启动，只会打印「警告: 唤醒词服务启动失败」：页面能打开，麦克风的输入流也已经打开，但没有唤醒词检测在消费这些数据，语音唤不醒。
+
 ### 修改唤醒词

 1. 打开数字人页面设置
@@ -91,6 +91,8 @@ TTS:
 ```py
 python app.py
 ```
+启动`main/digital-human`下的`python start.py`前，先按[唤醒词模型说明](./digital-human-wakeword.md)把模型下载并配置到`main/digital-human/wakeword_runtime/models/`，默认配置下缺模型只会打印「警告: 唤醒词服务启动失败」。页面能打开，麦克风的输入流也已经打开，但没有唤醒词检测在消费这些数据，语音唤不醒。
+
 启动`main/digital-human`下的`python start.py`后，打开`http://127.0.0.1:8006/index.html`，测试连接和发送消息时paddlespeech端是否有输出日志

 输出日志参考：
@@ -6,7 +6,7 @@
 Dieses Projekt basiert auf der Theorie und Technologie der Mensch-Maschine-symbiotischen Intelligenz zur Entwicklung intelligenter Terminal-Hardware- und Software-Systeme<br/>und bietet Backend-Dienste für das Open-Source-Hardware-Projekt
 <a href="https://github.com/78/xiaozhi-esp32">xiaozhi-esp32</a><br/>
 Implementiert mit Python, Java und Vue gemäß dem <a href="https://ccnphfhqs21z.feishu.cn/wiki/M0XiwldO9iJwHikpXD5cEx71nKh">Xiaozhi-Kommunikationsprotokoll</a><br/>
-Unterstützt MQTT+UDP-Protokoll, Websocket-Protokoll, MCP-Endpunkte, Stimmabdruckerkennung und Wissensdatenbank
+Unterstützt MQTT+UDP-Protokoll, Websocket-Protokoll, MCP-Endpunkte und Stimmabdruckerkennung
 </p>

 <p align="center">
@@ -55,31 +55,31 @@ Möchten Sie die Nutzungseffekte sehen? Klicken Sie auf die Videos unten 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1FMFyejExX" target="_blank">
        <picture>
-          <img alt="Gefühl der Antwortgeschwindigkeit" src="../images/demo9.png" /></picture>
+          <img alt="响应速度感受" src="../images/demo9.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1vchQzaEse" target="_blank">
        <picture>
-          <img alt="Das Geheimnis der Geschwindigkeitsoptimierung" src="../images/demo6.png" /></picture>
+          <img alt="速度优化秘诀" src="../images/demo6.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1WEcxzFEAT" target="_blank">
        <picture>
-          <img alt="Xiaozhi-Digitalmensch mit Sprachaktivierung" src="../images/demo8.png" /></picture>
+          <img alt="小智数字人 支持语音唤醒" src="../images/demo8.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1CKVz6UEuB" target="_blank">
        <picture>
-          <img alt="Anruf von Gerät zu Gerät" src="../images/demo0.png" /></picture>
+          <img alt="设备呼叫设备，打电话" src="../images/demo0.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1C1tCzUEZh" target="_blank">
        <picture>
-          <img alt="Komplexes medizinisches Szenario" src="../images/demo1.png" /></picture>
+          <img alt="复杂医疗场景" src="../images/demo1.png" /></picture>
      </a>
    </td>
  </tr>
@@ -87,31 +87,31 @@ Möchten Sie die Nutzungseffekte sehen? Klicken Sie auf die Videos unten 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1VC96Y5EMH" target="_blank">
        <picture>
-          <img alt="Musik abspielen, Wetter abfragen und Nachrichten vorlesen" src="../images/demo7.png" /></picture>
+          <img alt="播放音乐查询天气播报新闻" src="../images/demo7.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV12J7WzBEaH" target="_blank">
        <picture>
-          <img alt="Unterbrechung in Echtzeit" src="../images/demo10.png" /></picture>
+          <img alt="实时打断" src="../images/demo10.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1Co76z7EvK" target="_blank">
        <picture>
-          <img alt="Objekte per Foto erkennen" src="../images/demo12.png" /></picture>
+          <img alt="拍照识物品" src="../images/demo12.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1pNXWYGEx1" target="_blank">
        <picture>
-          <img alt="Haushaltsgeräte ein- und ausschalten" src="../images/demo5.png" /></picture>
+          <img alt="控制家电开关" src="../images/demo5.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1TJ7WzzEo6" target="_blank">
        <picture>
-          <img alt="Aufgaben mit mehreren Befehlen" src="../images/demo11.png" /></picture>
+          <img alt="多指令任务" src="../images/demo11.png" /></picture>
      </a>
    </td>
  </tr>
@@ -119,31 +119,31 @@ Möchten Sie die Nutzungseffekte sehen? Klicken Sie auf die Videos unten 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1ZQKUzYExM" target="_blank">
        <picture>
-          <img alt="MCP-Endpunkt" src="../images/demo13.png" /></picture>
+          <img alt="MCP接入点" src="../images/demo13.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1zUW5zJEkq" target="_blank">
        <picture>
-          <img alt="MQTT-Befehlsübermittlung" src="../images/demo4.png" /></picture>
+          <img alt="MQTT指令下发" src="../images/demo4.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1Exu3zqEDe" target="_blank">
        <picture>
-          <img alt="Stimmabdruckerkennung" src="../images/demo14.png" /></picture>
+          <img alt="声纹识别" src="../images/demo14.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1CDKWemEU6" target="_blank">
        <picture>
-          <img alt="Eigene Stimmfarbe" src="../images/demo2.png" /></picture>
+          <img alt="自定义音色" src="../images/demo2.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV12yA2egEaC" target="_blank">
        <picture>
-          <img alt="Gespräch auf Kantonesisch" src="../images/demo3.png" /></picture>
+          <img alt="使用粤语交流" src="../images/demo3.png" /></picture>
      </a>
    </td>
  </tr>
@@ -210,7 +210,7 @@ Dieses Projekt bietet die folgenden Testwerkzeuge, um Ihnen bei der Überprüfun

 | Werkzeugname | Standort | Verwendungsmethode | Funktionsbeschreibung |
 |:---:|:---|:---:|:---:|
-| Audio-Interaktionstesttool | main》digital-human》index.html | In `main/digital-human` zuerst die Abhängigkeiten wie in der README installieren, dann `python start.py` ausführen und `http://127.0.0.1:8006/index.html` öffnen | Testet Audio-Wiedergabe- und Empfangsfunktionen, überprüft, ob die Python-seitige Audioverarbeitung normal ist |
+| Audio-Interaktionstesttool | main》digital-human》index.html | In `main/digital-human` `python start.py` ausführen und dann `http://127.0.0.1:8006/index.html` öffnen | Testet Audio-Wiedergabe- und Empfangsfunktionen, überprüft, ob die Python-seitige Audioverarbeitung normal ist |
 | Modell-Reaktionstesttool | main》xiaozhi-server》performance_tester.py | Ausführen `python performance_tester.py` | Testet die Reaktionsgeschwindigkeit von Kernmodulen wie ASR (Spracherkennung), LLM (großes Modell), VLLM (Vision-Modell), TTS (Sprachsynthese) |

 > 💡 Hinweis: Beim Testen der Modellgeschwindigkeit werden nur Modelle mit konfigurierten Schlüsseln getestet.
@@ -218,7 +218,7 @@ Dieses Projekt bietet die folgenden Testwerkzeuge, um Ihnen bei der Überprüfun
 ---
 ## Funktionsliste ✨
 ### Implementiert ✅
-![Architekturdiagramm der Installation aller Module](../images/deploy2.png)
+![请参考-全模块安装架构图](../images/deploy2.png)
 | Funktionsmodul | Beschreibung |
 |:---:|:---|
 | Kernarchitektur | Basierend auf [MQTT+UDP-Gateway](https://github.com/xinnan-tech/xiaozhi-esp32-server/blob/main/docs/mqtt-gateway-integration.md), WebSocket und HTTP-Servern, bietet vollständiges Konsolenverwaltungs- und Authentifizierungssystem |
@@ -226,7 +226,7 @@ Dieses Projekt bietet die folgenden Testwerkzeuge, um Ihnen bei der Überprüfun
 | Stimmabdruckerkennung | Unterstützt Mehrbenutzer-Stimmabdruckregistrierung, -verwaltung und -erkennung, verarbeitet parallel mit ASR, Echtzeit-Sprecheridentitätserkennung und Weitergabe an LLM für personalisierte Antworten |
 | Intelligenter Dialog | Unterstützt mehrere LLM (große Sprachmodelle), implementiert intelligenten Dialog |
 | Visuelle Wahrnehmung | Unterstützt mehrere VLLM (Vision Large Models), implementiert multimodale Interaktion |
-| Absichtserkennung | Unterstützt externe LLM-Absichtserkennung, autonome Funktionsaufrufe durch große Modelle, bietet plugin-basierten Absichtsverarbeitungsmechanismus |
+| Absichtserkennung | Unterstützt LLM-Absichtserkennung, Function Call-Funktionsaufruf, bietet plugin-basierten Absichtsverarbeitungsmechanismus |
 | Gedächtnissystem | Unterstützt lokales Kurzzeitgedächtnis, mem0ai-Schnittstellengedächtnis, PowerMem intelligentes Gedächtnis, mit Gedächtniszusammenfassungsfunktion |
 | Wissensdatenbank | Unterstützt RAGFlow-Wissensdatenbank, ermöglicht großem Modell die Bewertung, ob Wissensdatenbank benötigt wird, bevor geantwortet wird |
 | Werkzeugaufruf | Unterstützt Client-IOT-Protokoll, Client-MCP-Protokoll, Server-MCP-Protokoll, MCP-Endpunktprotokoll, benutzerdefinierte Werkzeugfunktionen |
@@ -254,7 +254,7 @@ Xiaozhi ist ein Ökosystem. Wenn Sie dieses Produkt verwenden, können Sie sich

 | Verwendungsmethode | Unterstützte Plattformen | Kostenlose Plattformen |
 |:---:|:---:|:---:|
-| OpenAI-Schnittstellenaufrufe | Alibaba Bailian, Volcano Engine Doubao, Volcano Engine AI Gateway, DeepSeek, Zhipu ChatGLM, Gemini, iFLYTEK, LM Studio | Zhipu ChatGLM, Gemini |
+| OpenAI-Schnittstellenaufrufe | Alibaba Bailian, Volcano Engine Doubao, DeepSeek, Zhipu ChatGLM, Gemini, iFLYTEK | Zhipu ChatGLM, Gemini |
 | Ollama-Schnittstellenaufrufe | Ollama | - |
 | Dify-Schnittstellenaufrufe | Dify | - |
 | FastGPT-Schnittstellenaufrufe | FastGPT | - |
@@ -280,7 +280,7 @@ Tatsächlich kann jedes VLLM, das OpenAI-Schnittstellenaufrufe unterstützt, int

 | Verwendungsmethode | Unterstützte Plattformen | Kostenlose Plattformen |
 |:---:|:---:|:---:|
-| Schnittstellenaufrufe | EdgeTTS, iFLYTEK, Volcano Engine Doubao TTS, Volcano Engine AI Gateway TTS, Tencent Cloud, Alibaba Cloud TTS, AliYun Stream TTS, Alibaba Bailian CosyVoice TTS, CosyVoiceSiliconflow, TTS302AI, CozeCnTTS, CustomTTS, OpenAITTS, MinimaxTTS, Volcano Dual-Stream TTS | EdgeTTS, CosyVoiceSiliconflow (teilweise) |
+| Schnittstellenaufrufe | EdgeTTS, iFLYTEK, Volcano Engine Doubao TTS, Tencent Cloud, Alibaba Cloud TTS, AliYun Stream TTS, CosyVoiceSiliconflow, TTS302AI, CozeCnTTS, CustomTTS, OpenAITTS, MinimaxTTS, Volcano Dual-Stream TTS | EdgeTTS, CosyVoiceSiliconflow (teilweise) |
 | Lokale Dienste | FishSpeech, GPT_SOVITS_V2, GPT_SOVITS_V3, IndexStreamTTS, PaddleSpeechTTS | IndexStreamTTS, PaddleSpeechTTS, FishSpeech, GPT_SOVITS_V2, GPT_SOVITS_V3 |

 ---
@@ -298,7 +298,7 @@ Tatsächlich kann jedes VLLM, das OpenAI-Schnittstellenaufrufe unterstützt, int
 | Verwendungsmethode | Unterstützte Plattformen | Kostenlose Plattformen |
 |:---:|:---:|:---:|
 | Lokale Verwendung | FunASR, SherpaASR, SherpaParaformerASR, VoskASR | FunASR, SherpaASR, SherpaParaformerASR, VoskASR |
-| Schnittstellenaufrufe | DoubaoASR, Doubao Streaming ASR, Doubao Streaming ASR V2, FunASRServer, iFLYTEK, TencentASR, AliyunASR, Aliyun Streaming ASR, Alibaba Bailian Streaming ASR, Baidu ASR, OpenAI ASR, GroqASR, Qwen3ASRFlash | FunASRServer |
+| Schnittstellenaufrufe | DoubaoASR, Doubao Streaming ASR, FunASRServer, iFLYTEK, TencentASR, AliyunASR, Aliyun Streaming ASR, Baidu ASR, OpenAI ASR, GroqASR, Qwen3ASRFlash | FunASRServer |

 ---

@@ -354,10 +354,11 @@ Die Stimmabdruckerkennung erfordert einen separat bereitgestellten [voiceprint-a
 | <img src="../images/logo_contributors.png" width="160"> | [Code-Mitwirkende](https://github.com/xinnan-tech/xiaozhi-esp32-server/graphs/contributors) | Dank an [alle Code-Mitwirkenden](https://github.com/xinnan-tech/xiaozhi-esp32-server/graphs/contributors), Ihre Bemühungen haben das Projekt robuster und leistungsfähiger gemacht. |


-<a href="https://www.star-history.com/?repos=xinnan-tech%2Fxiaozhi-esp32-server&type=date&legend=top-left">
+<a href="https://star-history.com/#xinnan-tech/xiaozhi-esp32-server&Date">
+
 <picture>
-   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&theme=dark&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
-   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
-   <img alt="Star History Chart" src="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
+   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date&theme=dark" />
+   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date" />
+   <img alt="Star History Chart" src="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date" />
 </picture>
 </a>
@@ -55,31 +55,31 @@ Want to see the usage effects? Click the videos below 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1FMFyejExX" target="_blank">
        <picture>
-          <img alt="The feel of the response speed" src="../images/demo9.png" /></picture>
+          <img alt="响应速度感受" src="../images/demo9.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1vchQzaEse" target="_blank">
        <picture>
-          <img alt="The secret to speeding things up" src="../images/demo6.png" /></picture>
+          <img alt="速度优化秘诀" src="../images/demo6.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1WEcxzFEAT" target="_blank">
        <picture>
-          <img alt="Xiaozhi digital human with voice wake-up" src="../images/demo8.png" /></picture>
+          <img alt="小智数字人 支持语音唤醒" src="../images/demo8.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1CKVz6UEuB" target="_blank">
        <picture>
-          <img alt="Device-to-device calling" src="../images/demo0.png" /></picture>
+          <img alt="设备呼叫设备，打电话" src="../images/demo0.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1C1tCzUEZh" target="_blank">
        <picture>
-          <img alt="A complex medical scenario" src="../images/demo1.png" /></picture>
+          <img alt="复杂医疗场景" src="../images/demo1.png" /></picture>
      </a>
    </td>
  </tr>
@@ -87,31 +87,31 @@ Want to see the usage effects? Click the videos below 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1VC96Y5EMH" target="_blank">
        <picture>
-          <img alt="Play music, check the weather, and read the news" src="../images/demo7.png" /></picture>
+          <img alt="播放音乐查询天气播报新闻" src="../images/demo7.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV12J7WzBEaH" target="_blank">
        <picture>
-          <img alt="Real-time interruption" src="../images/demo10.png" /></picture>
+          <img alt="实时打断" src="../images/demo10.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1Co76z7EvK" target="_blank">
        <picture>
-          <img alt="Take a photo to identify objects" src="../images/demo12.png" /></picture>
+          <img alt="拍照识物品" src="../images/demo12.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1pNXWYGEx1" target="_blank">
        <picture>
-          <img alt="Turn home appliances on and off" src="../images/demo5.png" /></picture>
+          <img alt="控制家电开关" src="../images/demo5.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1TJ7WzzEo6" target="_blank">
        <picture>
-          <img alt="Multi-command tasks" src="../images/demo11.png" /></picture>
+          <img alt="多指令任务" src="../images/demo11.png" /></picture>
      </a>
    </td>
  </tr>
@@ -119,31 +119,31 @@ Want to see the usage effects? Click the videos below 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1ZQKUzYExM" target="_blank">
        <picture>
-          <img alt="MCP access point" src="../images/demo13.png" /></picture>
+          <img alt="MCP接入点" src="../images/demo13.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1zUW5zJEkq" target="_blank">
        <picture>
-          <img alt="MQTT command delivery" src="../images/demo4.png" /></picture>
+          <img alt="MQTT指令下发" src="../images/demo4.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1Exu3zqEDe" target="_blank">
        <picture>
-          <img alt="Voiceprint recognition" src="../images/demo14.png" /></picture>
+          <img alt="声纹识别" src="../images/demo14.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1CDKWemEU6" target="_blank">
        <picture>
-          <img alt="Custom voice timbre" src="../images/demo2.png" /></picture>
+          <img alt="自定义音色" src="../images/demo2.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV12yA2egEaC" target="_blank">
        <picture>
-          <img alt="Talking in Cantonese" src="../images/demo3.png" /></picture>
+          <img alt="使用粤语交流" src="../images/demo3.png" /></picture>
      </a>
    </td>
  </tr>
@@ -210,7 +210,7 @@ This project provides the following testing tools to help you verify the system

 | Tool Name | Location | Usage Method | Function Description |
 |:---:|:---|:---:|:---:|
-| Audio Interaction Test Tool | main》digital-human》index.html | In `main/digital-human`, install dependencies as its README says, then run `python start.py` and open `http://127.0.0.1:8006/index.html` | Tests audio playback and reception functions, verifies if Python-side audio processing is normal |
+| Audio Interaction Test Tool | main》digital-human》index.html | Run `python start.py` in `main/digital-human`, then open `http://127.0.0.1:8006/index.html` | Tests audio playback and reception functions, verifies if Python-side audio processing is normal |
 | Model Response Test Tool | main》xiaozhi-server》performance_tester.py | Execute `python performance_tester.py` | Tests response speed of core modules such as ASR(speech recognition), LLM(large model), VLLM(vision model), TTS(speech synthesis) |

 > 💡 Note: When testing model speed, only models with configured keys will be tested.
@@ -218,7 +218,7 @@ This project provides the following testing tools to help you verify the system
 ---
 ## Feature List ✨
 ### Implemented ✅
-![Full-module installation architecture diagram](../images/deploy2.png)
+![请参考-全模块安装架构图](../images/deploy2.png)
 | Feature Module | Description |
 |:---:|:---|
 | Core Architecture | Based on [MQTT+UDP gateway](https://github.com/xinnan-tech/xiaozhi-esp32-server/blob/main/docs/mqtt-gateway-integration.md), WebSocket and HTTP servers, provides complete console management and authentication system |
@@ -226,7 +226,7 @@ This project provides the following testing tools to help you verify the system
 | Voiceprint Recognition | Supports multi-user voiceprint registration, management, and recognition, processes in parallel with ASR, real-time speaker identity recognition and passes to LLM for personalized responses |
 | Intelligent Dialogue | Supports multiple LLM(large language models), implements intelligent dialogue |
 | Visual Perception | Supports multiple VLLM(vision large models), implements multimodal interaction |
-| Intent Recognition | Supports pluggable external LLM intent recognition and autonomous LLM function calling, provides plugin-based intent processing mechanism |
+| Intent Recognition | Supports LLM intent recognition, Function Call function calling, provides plugin-based intent processing mechanism |
 | Memory System | Supports local short-term memory, mem0ai interface memory, PowerMem intelligent memory, with memory summarization functionality |
 | Knowledge Base | Supports RAGFlow knowledge base, enabling LLM to judge whether to schedule the knowledge base after receiving the user's question, and then answer the question |
 | Tool Calling | Supports client IOT protocol, client MCP protocol, server MCP protocol, MCP endpoint protocol, custom tool functions |
@@ -254,7 +254,7 @@ Xiaozhi is an ecosystem. When using this product, you can also check out other [

 | Usage Method | Supported Platforms | Free Platforms |
 |:---:|:---:|:---:|
-| OpenAI interface calls | Alibaba Bailian, Volcano Engine, DeepSeek, Zhipu, Gemini, iFLYTEK, LM Studio | Zhipu, Gemini |
+| OpenAI interface calls | Alibaba Bailian, Volcano Engine, DeepSeek, Zhipu, Gemini, iFLYTEK | Zhipu, Gemini |
 | Ollama interface calls | Ollama | - |
 | Dify interface calls | Dify | - |
 | FastGPT interface calls | FastGPT | - |
@@ -298,7 +298,7 @@ In fact, any VLLM that supports OpenAI interface calls can be integrated and use
 | Usage Method | Supported Platforms | Free Platforms |
 |:---:|:---:|:---:|
 | Local use | FunASR, SherpaASR, SherpaParaformerASR, VoskASR | FunASR, SherpaASR, SherpaParaformerASR, VoskASR |
-| Interface calls | FunASRServer, Volcano Engine, iFLYTEK, Tencent Cloud, Alibaba Cloud and Bailian, Baidu Cloud, OpenAI ASR, GroqASR, Qwen3ASRFlash | FunASRServer |
+| Interface calls | FunASRServer, Volcano Engine, iFLYTEK, Tencent Cloud, Alibaba Cloud, Baidu Cloud, OpenAI ASR, GroqASR, Qwen3ASRFlash | FunASRServer |

 ---

@@ -354,10 +354,11 @@ Voiceprint recognition requires a separately deployed [voiceprint-api](https://g
 | <img src="../images/logo_contributors.png" width="160"> | [Code Contributors](https://github.com/xinnan-tech/xiaozhi-esp32-server/graphs/contributors) | Thanks to [all code contributors](https://github.com/xinnan-tech/xiaozhi-esp32-server/graphs/contributors), your efforts have made the project more robust and powerful. |


-<a href="https://www.star-history.com/?repos=xinnan-tech%2Fxiaozhi-esp32-server&type=date&legend=top-left">
+<a href="https://star-history.com/#xinnan-tech/xiaozhi-esp32-server&Date">
+
 <picture>
-   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&theme=dark&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
-   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
-   <img alt="Star History Chart" src="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
+   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date&theme=dark" />
+   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date" />
+   <img alt="Star History Chart" src="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date" />
 </picture>
 </a>
@@ -168,8 +168,8 @@ Este proyecto ofrece dos formas de despliegue. Elija según lo que necesite:
 #### 🚀 Elección del método de despliegue
 | Método de despliegue | Características | Escenarios | Documentación de despliegue | Requisitos | Tutoriales en vídeo |
 |---------|------|---------|---------|---------|---------|
-| **Instalación simplificada** | Diálogo inteligente, gestión de un solo agente | Entornos con pocos recursos; los datos se guardan en archivos de configuración y no hace falta base de datos | [①Versión Docker](../Deployment.md#%E6%96%B9%E5%BC%8F%E4%B8%80docker%E5%8F%AA%E8%BF%90%E8%A1%8Cserver) / [②Despliegue desde el código fuente](../Deployment.md#%E6%96%B9%E5%BC%8F%E4%BA%8C%E6%9C%AC%E5%9C%B0%E6%BA%90%E7%A0%81%E5%8F%AA%E8%BF%90%E8%A1%8Cserver)| Ejecutar `FunASR` en local consume bastante memoria; si todo va por API, el consumo de recursos es bajo | - |
-| **Instalación de todos los módulos** | Diálogo inteligente, gestión multiusuario, gestión de varios agentes, uso de la interfaz de la consola inteligente | Experiencia con todas las funciones; los datos se guardan en una base de datos |[①Versión Docker](../Deployment_all.md#%E6%96%B9%E5%BC%8F%E4%B8%80docker%E8%BF%90%E8%A1%8C%E5%85%A8%E6%A8%A1%E5%9D%97) / [②Despliegue desde el código fuente](../Deployment_all.md#%E6%96%B9%E5%BC%8F%E4%BA%8C%E6%9C%AC%E5%9C%B0%E6%BA%90%E7%A0%81%E8%BF%90%E8%A1%8C%E5%85%A8%E6%A8%A1%E5%9D%97) / [③Tutorial de actualización automática desde el código fuente](../dev-ops-integration.md) | Ejecutar `FunASR` en local consume bastante memoria, y además requiere recursos de base de datos; si todo va por API, el consumo de recursos es bajo | [Tutorial en vídeo para arrancar el código fuente en local](https://www.bilibili.com/video/BV1wBJhz4Ewe) |
+| **Instalación simplificada** | Diálogo inteligente, gestión de un solo agente | Entornos con pocos recursos; los datos se guardan en archivos de configuración y no hace falta base de datos | [①Versión Docker](../Deployment.md#%E6%96%B9%E5%BC%8F%E4%B8%80docker%E5%8F%AA%E8%BF%90%E8%A1%8Cserver) / [②Despliegue desde el código fuente](../Deployment.md#%E6%96%B9%E5%BC%8F%E4%BA%8C%E6%9C%AC%E5%9C%B0%E6%BA%90%E7%A0%81%E5%8F%AA%E8%BF%90%E8%A1%8Cserver)| 2 núcleos y 4 GB si usa `FunASR`; 2 núcleos y 2 GB si todo va por API | - |
+| **Instalación de todos los módulos** | Diálogo inteligente, gestión multiusuario, gestión de varios agentes, uso de la interfaz de la consola inteligente | Experiencia con todas las funciones; los datos se guardan en una base de datos |[①Versión Docker](../Deployment_all.md#%E6%96%B9%E5%BC%8F%E4%B8%80docker%E8%BF%90%E8%A1%8C%E5%85%A8%E6%A8%A1%E5%9D%97) / [②Despliegue desde el código fuente](../Deployment_all.md#%E6%96%B9%E5%BC%8F%E4%BA%8C%E6%9C%AC%E5%9C%B0%E6%BA%90%E7%A0%81%E8%BF%90%E8%A1%8C%E5%85%A8%E6%A8%A1%E5%9D%97) / [③Tutorial de actualización automática desde el código fuente](../dev-ops-integration.md) | 4 núcleos y 8 GB si usa `FunASR`; 2 núcleos y 4 GB si todo va por API | [Tutorial en vídeo para arrancar el código fuente en local](https://www.bilibili.com/video/BV1wBJhz4Ewe) |

 Las preguntas frecuentes y los tutoriales relacionados están en [este enlace](../FAQ.md)

@@ -210,8 +210,8 @@ Este proyecto incluye las siguientes herramientas para comprobar el sistema y el

 | Herramienta | Ubicación | Uso | Qué hace |
 |:---:|:---|:---:|:---:|
-| Prueba de interacción de audio | main》digital-human》index.html | En `main/digital-human`, instale las dependencias como indica su README, después ejecute `python start.py` y abra `http://127.0.0.1:8006/index.html` | Prueba la reproducción y la recepción de audio, y comprueba que el procesamiento de audio en Python funciona |
-| Prueba de respuesta de los modelos | main》xiaozhi-server》performance_tester.py | Ejecute `python performance_tester.py` | Mide la velocidad de respuesta de los principales módulos: ASR (reconocimiento de voz), LLM (modelo de lenguaje), VLLM (modelo de visión) y TTS (síntesis de voz) |
+| Prueba de interacción de audio | main》digital-human》index.html | En `main/digital-human`, ejecute `python start.py` y abra `http://127.0.0.1:8006/index.html` | Prueba la reproducción y la recepción de audio, y comprueba que el procesamiento de audio en Python funciona |
+| Prueba de respuesta de los modelos | main》xiaozhi-server》performance_tester.py | Ejecute `python performance_tester.py` | Mide la velocidad de respuesta de los cuatro módulos principales: ASR (reconocimiento de voz), LLM (modelo de lenguaje), VLLM (modelo de visión) y TTS (síntesis de voz) |

 > 💡 Nota: al medir la velocidad de los modelos, solo se prueban los modelos que tienen una clave configurada.

@@ -226,15 +226,15 @@ Este proyecto incluye las siguientes herramientas para comprobar el sistema y el
 | Huella de voz | Admite el registro, la gestión y el reconocimiento de la huella de voz de varios usuarios; se procesa en paralelo con el ASR, identifica en tiempo real quién habla y se lo pasa al LLM para una respuesta personalizada |
 | Diálogo inteligente | Admite varios LLM (modelos de lenguaje) para mantener un diálogo inteligente |
 | Percepción visual | Admite varios VLLM (modelos de visión) para una interacción multimodal |
-| Reconocimiento de intención | Admite el reconocimiento de intención con un modelo de lenguaje (LLM) externo y conectable, y la llamada autónoma a funciones del modelo, y ofrece un mecanismo de procesamiento de intenciones mediante plugins |
+| Reconocimiento de intención | Admite el reconocimiento de intención con el modelo de lenguaje (LLM) y la llamada autónoma a funciones del modelo, y ofrece un mecanismo de procesamiento de intenciones mediante plugins |
 | Sistema de memoria | Admite memoria local a corto plazo, memoria por la interfaz mem0ai y memoria inteligente PowerMem, con resumen de la memoria |
 | Base de conocimiento | Admite la base de conocimiento RAGFlow, para que el modelo decida si debe consultarla antes de responder |
 | Llamada a herramientas | Admite el protocolo IOT del cliente, el protocolo MCP del cliente, el protocolo MCP del servidor, el protocolo de punto de acceso MCP y funciones de herramientas personalizadas |
 | Envío de instrucciones | Apoyado en el protocolo MQTT, permite enviar instrucciones MCP desde la consola inteligente a los dispositivos ESP32 |
-| Backend de administración | Ofrece una interfaz web de administración, con gestión de usuarios, configuración del sistema y gestión de dispositivos. La interfaz se muestra en chino simplificado, chino tradicional, inglés, alemán, vietnamita y portugués de Brasil |
-| Herramientas de prueba | Incluye herramientas de prueba de rendimiento y de interacción de audio |
+| Backend de administración | Ofrece una interfaz web de administración, con gestión de usuarios, configuración del sistema y gestión de dispositivos. La interfaz se muestra en chino simplificado, chino tradicional e inglés |
+| Herramientas de prueba | Incluye herramientas de prueba de rendimiento, de modelos de visión y de interacción de audio |
 | Soporte de despliegue | Admite el despliegue con Docker y el despliegue local, y ofrece una gestión completa de los archivos de configuración |
-| Sistema de plugins | Admite extensiones de plugins funcionales y el desarrollo de plugins propios; los nuevos archivos de plugin se cargan automáticamente en la siguiente conexión del dispositivo, y los cambios en el código de un plugin ya cargado requieren reiniciar el servicio |
+| Sistema de plugins | Admite extensiones de plugins funcionales, el desarrollo de plugins propios y la carga en caliente de plugins |

 ### En desarrollo 🚧

@@ -254,7 +254,7 @@ Xiaozhi es un ecosistema. Al usar este producto, también puede ver otros [proye

 | Forma de uso | Plataformas compatibles | Plataformas gratuitas |
 |:---:|:---:|:---:|
-| Llamadas por la interfaz OpenAI | Alibaba Bailian, Volcano Engine, DeepSeek, Zhipu, Gemini, iFLYTEK, LM Studio | Zhipu, Gemini |
+| Llamadas por la interfaz OpenAI | Alibaba Bailian, Volcano Engine, DeepSeek, Zhipu, Gemini, iFLYTEK | Zhipu, Gemini |
 | Llamadas por la interfaz Ollama | Ollama | - |
 | Llamadas por la interfaz Dify | Dify | - |
 | Llamadas por la interfaz FastGPT | FastGPT | - |
@@ -280,8 +280,8 @@ En la práctica, se puede integrar cualquier VLLM que admita llamadas por la int

 | Forma de uso | Plataformas compatibles | Plataformas gratuitas |
 |:---:|:---:|:---:|
-| Llamadas por interfaz | EdgeTTS, iFLYTEK, Volcano Engine, Tencent Cloud, Alibaba Cloud y Bailian, CosyVoiceSiliconflow, TTS302AI, CozeCnTTS, CustomTTS, OpenAITTS, MinimaxTTS | EdgeTTS, CosyVoiceSiliconflow (parcial) |
-| Servicio local | FishSpeech, GPT_SOVITS_V2, GPT_SOVITS_V3, IndexStreamTTS, PaddleSpeechTTS | IndexStreamTTS, PaddleSpeechTTS, FishSpeech, GPT_SOVITS_V2, GPT_SOVITS_V3 |
+| Llamadas por interfaz | EdgeTTS, iFLYTEK, Volcano Engine, Tencent Cloud, Alibaba Cloud y Bailian, CosyVoiceSiliconflow, TTS302AI, CozeCnTTS, GizwitsTTS, ACGNTTS, OpenAITTS, Lingxi Streaming TTS, MinimaxTTS | Lingxi Streaming TTS, EdgeTTS, CosyVoiceSiliconflow (parcial) |
+| Servicio local | FishSpeech, GPT_SOVITS_V2, GPT_SOVITS_V3, Index-TTS, PaddleSpeech | Index-TTS, PaddleSpeech, FishSpeech, GPT_SOVITS_V2, GPT_SOVITS_V3 |

 ---

@@ -297,8 +297,8 @@ En la práctica, se puede integrar cualquier VLLM que admita llamadas por la int

 | Forma de uso | Plataformas compatibles | Plataformas gratuitas |
 |:---:|:---:|:---:|
-| Uso local | FunASR, SherpaASR, SherpaParaformerASR, VoskASR | FunASR, SherpaASR, SherpaParaformerASR, VoskASR |
-| Llamadas por interfaz | FunASRServer, Volcano Engine, iFLYTEK, Tencent Cloud, Alibaba Cloud y Bailian, Baidu Cloud, OpenAI ASR, GroqASR, Qwen3ASRFlash | FunASRServer |
+| Uso local | FunASR, SherpaASR | FunASR, SherpaASR |
+| Llamadas por interfaz | FunASRServer, Volcano Engine, iFLYTEK, Tencent Cloud, Alibaba Cloud, Baidu Cloud, OpenAI ASR | FunASRServer |

 ---

@@ -306,9 +306,7 @@ En la práctica, se puede integrar cualquier VLLM que admita llamadas por la int

 | Forma de uso | Plataformas compatibles | Plataformas gratuitas |
 |:---:|:---:|:---:|
-| Llamadas por interfaz | Servicio de reconocimiento de huella de voz (voiceprint-api) | - |
-
-El reconocimiento de huella de voz requiere desplegar por separado el servicio [voiceprint-api](https://github.com/xinnan-tech/voiceprint-api) (basado en 3D-Speaker), que depende de MySQL; consulte la [guía de integración de la huella de voz](../voiceprint-integration.md)
+| Uso local | 3D-Speaker | 3D-Speaker |

 ---

@@ -55,31 +55,31 @@ Quer ver os efeitos de uso? Clique nos vídeos abaixo 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1FMFyejExX" target="_blank">
        <picture>
-          <img alt="A sensação da velocidade de resposta" src="../images/demo9.png" /></picture>
+          <img alt="响应速度感受" src="../images/demo9.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1vchQzaEse" target="_blank">
        <picture>
-          <img alt="O segredo para otimizar a velocidade" src="../images/demo6.png" /></picture>
+          <img alt="速度优化秘诀" src="../images/demo6.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1WEcxzFEAT" target="_blank">
        <picture>
-          <img alt="Humano digital Xiaozhi com ativação por voz" src="../images/demo8.png" /></picture>
+          <img alt="小智数字人 支持语音唤醒" src="../images/demo8.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1CKVz6UEuB" target="_blank">
        <picture>
-          <img alt="Chamadas de dispositivo para dispositivo" src="../images/demo0.png" /></picture>
+          <img alt="设备呼叫设备，打电话" src="../images/demo0.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1C1tCzUEZh" target="_blank">
        <picture>
-          <img alt="Cenário médico complexo" src="../images/demo1.png" /></picture>
+          <img alt="复杂医疗场景" src="../images/demo1.png" /></picture>
      </a>
    </td>
  </tr>
@@ -87,31 +87,31 @@ Quer ver os efeitos de uso? Clique nos vídeos abaixo 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1VC96Y5EMH" target="_blank">
        <picture>
-          <img alt="Reproduzir música, consultar o tempo e ler as notícias" src="../images/demo7.png" /></picture>
+          <img alt="播放音乐查询天气播报新闻" src="../images/demo7.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV12J7WzBEaH" target="_blank">
        <picture>
-          <img alt="Interrupção em tempo real" src="../images/demo10.png" /></picture>
+          <img alt="实时打断" src="../images/demo10.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1Co76z7EvK" target="_blank">
        <picture>
-          <img alt="Reconhecer objetos com uma foto" src="../images/demo12.png" /></picture>
+          <img alt="拍照识物品" src="../images/demo12.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1pNXWYGEx1" target="_blank">
        <picture>
-          <img alt="Controlar o acionamento de eletrodomésticos" src="../images/demo5.png" /></picture>
+          <img alt="控制家电开关" src="../images/demo5.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1TJ7WzzEo6" target="_blank">
        <picture>
-          <img alt="Tarefas com vários comandos" src="../images/demo11.png" /></picture>
+          <img alt="多指令任务" src="../images/demo11.png" /></picture>
      </a>
    </td>
  </tr>
@@ -119,31 +119,31 @@ Quer ver os efeitos de uso? Clique nos vídeos abaixo 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1ZQKUzYExM" target="_blank">
        <picture>
-          <img alt="Ponto de acesso MCP" src="../images/demo13.png" /></picture>
+          <img alt="MCP接入点" src="../images/demo13.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1zUW5zJEkq" target="_blank">
        <picture>
-          <img alt="Envio de comandos MQTT" src="../images/demo4.png" /></picture>
+          <img alt="MQTT指令下发" src="../images/demo4.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1Exu3zqEDe" target="_blank">
        <picture>
-          <img alt="Reconhecimento de impressão vocal" src="../images/demo14.png" /></picture>
+          <img alt="声纹识别" src="../images/demo14.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1CDKWemEU6" target="_blank">
        <picture>
-          <img alt="Tom de voz personalizado" src="../images/demo2.png" /></picture>
+          <img alt="自定义音色" src="../images/demo2.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV12yA2egEaC" target="_blank">
        <picture>
-          <img alt="Conversa em cantonês" src="../images/demo3.png" /></picture>
+          <img alt="使用粤语交流" src="../images/demo3.png" /></picture>
      </a>
    </td>
  </tr>
@@ -210,7 +210,7 @@ Este projeto fornece as seguintes ferramentas de teste para ajudá-lo a verifica

 | Nome da Ferramenta | Localização | Método de Uso | Descrição da Função |
 |:---:|:---|:---:|:---:|
-| Ferramenta de Teste de Interação por Áudio | main》digital-human》index.html | Em `main/digital-human`, instale as dependências como indica o README, depois execute `python start.py` e abra `http://127.0.0.1:8006/index.html` | Testa as funções de reprodução e recepção de áudio, verifica se o processamento de áudio no lado Python está normal |
+| Ferramenta de Teste de Interação por Áudio | main》digital-human》index.html | Execute `python start.py` em `main/digital-human` e depois abra `http://127.0.0.1:8006/index.html` | Testa as funções de reprodução e recepção de áudio, verifica se o processamento de áudio no lado Python está normal |
 | Ferramenta de Teste de Resposta de Modelo | main》xiaozhi-server》performance_tester.py | Execute `python performance_tester.py` | Testa a velocidade de resposta de módulos principais como ASR(reconhecimento de fala), LLM(modelo de linguagem), VLLM(modelo de visão), TTS(síntese de voz) |

 > 💡 Nota: Ao testar a velocidade dos modelos, apenas os modelos com chaves configuradas serão testados.
@@ -218,7 +218,7 @@ Este projeto fornece as seguintes ferramentas de teste para ajudá-lo a verifica
 ---
 ## Lista de Funcionalidades ✨
 ### Implementado ✅
-![Diagrama de arquitetura da instalação de todos os módulos](../images/deploy2.png)
+![请参考-全模块安装架构图](../images/deploy2.png)
 | Módulo de Funcionalidade | Descrição |
 |:---:|:---|
 | Arquitetura Principal | Baseado em [gateway MQTT+UDP](https://github.com/xinnan-tech/xiaozhi-esp32-server/blob/main/docs/mqtt-gateway-integration.md), servidores WebSocket e HTTP, fornece sistema completo de gerenciamento de console e autenticação |
@@ -226,7 +226,7 @@ Este projeto fornece as seguintes ferramentas de teste para ajudá-lo a verifica
 | Reconhecimento de Impressão Vocal | Suporta registro, gerenciamento e reconhecimento de impressão vocal de múltiplos usuários, processa em paralelo com o ASR, reconhecimento de identidade do falante em tempo real e repassa ao LLM para respostas personalizadas |
 | Diálogo Inteligente | Suporta múltiplos LLM (modelos de linguagem de grande porte), implementa diálogo inteligente |
 | Percepção Visual | Suporta múltiplos VLLM (modelos de visão de grande porte), implementa interação multimodal |
-| Reconhecimento de Intenção | Suporta reconhecimento de intenção por LLM externo e conectável, chamada de função autônoma do modelo de linguagem, fornece mecanismo de processamento de intenção baseado em plugins |
+| Reconhecimento de Intenção | Suporta reconhecimento de intenção por LLM, Function Call (chamada de função), fornece mecanismo de processamento de intenção baseado em plugins |
 | Sistema de Memória | Suporta memória local de curto prazo, memória via interface mem0ai, memória inteligente PowerMem, com funcionalidade de resumo de memória |
 | Base de Conhecimento | Suporta base de conhecimento RAGFlow, permitindo que o LLM julgue se deve acionar a base de conhecimento após receber a pergunta do usuário, e então responda à pergunta |
 | Chamada de Ferramentas | Suporta protocolo IOT do cliente, protocolo MCP do cliente, protocolo MCP do servidor, protocolo de endpoint MCP, funções de ferramentas personalizadas |
@@ -254,7 +254,7 @@ Xiaozhi é um ecossistema. Ao utilizar este produto, você também pode conferir

 | Método de Uso | Plataformas Suportadas | Plataformas Gratuitas |
 |:---:|:---:|:---:|
-| Chamadas via interface OpenAI | Alibaba Bailian, Volcano Engine, DeepSeek, Zhipu, Gemini, iFLYTEK, LM Studio | Zhipu, Gemini |
+| Chamadas via interface OpenAI | Alibaba Bailian, Volcano Engine, DeepSeek, Zhipu, Gemini, iFLYTEK | Zhipu, Gemini |
 | Chamadas via interface Ollama | Ollama | - |
 | Chamadas via interface Dify | Dify | - |
 | Chamadas via interface FastGPT | FastGPT | - |
@@ -298,7 +298,7 @@ Na verdade, qualquer VLLM que suporte chamadas via interface OpenAI pode ser int
 | Método de Uso | Plataformas Suportadas | Plataformas Gratuitas |
 |:---:|:---:|:---:|
 | Uso local | FunASR, SherpaASR, SherpaParaformerASR, VoskASR | FunASR, SherpaASR, SherpaParaformerASR, VoskASR |
-| Chamadas via interface | FunASRServer, Volcano Engine, iFLYTEK, Tencent Cloud, Alibaba Cloud e Bailian, Baidu Cloud, OpenAI ASR, GroqASR, Qwen3ASRFlash | FunASRServer |
+| Chamadas via interface | FunASRServer, Volcano Engine, iFLYTEK, Tencent Cloud, Alibaba Cloud, Baidu Cloud, OpenAI ASR, GroqASR, Qwen3ASRFlash | FunASRServer |

 ---

@@ -308,7 +308,7 @@ Na verdade, qualquer VLLM que suporte chamadas via interface OpenAI pode ser int
 |:---:|:---:|:---:|
 | Chamadas via interface | Serviço de voiceprint (voiceprint-api) | - |

-O reconhecimento de impressão vocal exige um serviço [voiceprint-api](https://github.com/xinnan-tech/voiceprint-api) implantado separadamente (baseado em 3D-Speaker) e MySQL. Veja o [guia de integração de voiceprint](../voiceprint-integration.md).
+O reconhecimento de voz por impressão digital exige um serviço [voiceprint-api](https://github.com/xinnan-tech/voiceprint-api) implantado separadamente (baseado em 3D-Speaker) e MySQL. Veja o [guia de integração de voiceprint](../voiceprint-integration.md).

 ---

@@ -354,10 +354,11 @@ O reconhecimento de impressão vocal exige um serviço [voiceprint-api](https://
 | <img src="../images/logo_contributors.png" width="160"> | [Contribuidores de Código](https://github.com/xinnan-tech/xiaozhi-esp32-server/graphs/contributors) | Agradecimentos a [todos os contribuidores de código](https://github.com/xinnan-tech/xiaozhi-esp32-server/graphs/contributors), seus esforços tornaram o projeto mais robusto e poderoso. |


-<a href="https://www.star-history.com/?repos=xinnan-tech%2Fxiaozhi-esp32-server&type=date&legend=top-left">
+<a href="https://star-history.com/#xinnan-tech/xiaozhi-esp32-server&Date">
+
 <picture>
-   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&theme=dark&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
-   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
-   <img alt="Star History Chart" src="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
+   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date&theme=dark" />
+   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date" />
+   <img alt="Star History Chart" src="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date" />
 </picture>
 </a>
@@ -6,7 +6,7 @@
 Dự án này dựa trên lý thuyết và công nghệ trí tuệ cộng sinh người-máy để phát triển hệ thống phần mềm và phần cứng thiết bị đầu cuối thông minh<br/>Cung cấp dịch vụ backend cho dự án phần cứng thông minh mã nguồn mở
 <a href="https://github.com/78/xiaozhi-esp32">xiaozhi-esp32</a><br/>
 Được triển khai bằng Python, Java, Vue theo <a href="https://ccnphfhqs21z.feishu.cn/wiki/M0XiwldO9iJwHikpXD5cEx71nKh">giao thức truyền thông Xiaozhi</a><br/>
-Hỗ trợ giao thức MQTT+UDP, giao thức Websocket, điểm truy cập MCP, nhận dạng vân giọng và kho tri thức
+Hỗ trợ giao thức MQTT+UDP, giao thức Websocket, điểm truy cập MCP, nhận dạng giọng nói và kho tri thức
 </p>

 <p align="center">
@@ -39,7 +39,7 @@ Spearheaded by Professor Siyuan Liu's Team (South China University of Technology
 </br>
 Được dẫn dắt bởi nhóm Giáo sư Lưu Tư Nguyên (Đại học Bách khoa Nam Trung Quốc)
 </br>
-<img src="../images/hnlg.jpg" alt="Đại học Bách khoa Nam Trung Quốc" width="50%">
+<img src="../images/hnlg.jpg" alt="华南理工大学" width="50%">
 </p>

 ---
@@ -55,31 +55,31 @@ Muốn xem hiệu quả sử dụng? Hãy xem video 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1FMFyejExX" target="_blank">
        <picture>
-          <img alt="Cảm giác tốc độ phản hồi" src="../images/demo9.png" /></picture>
+          <img alt="响应速度感受" src="../images/demo9.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1vchQzaEse" target="_blank">
        <picture>
-          <img alt="Bí quyết tối ưu tốc độ" src="../images/demo6.png" /></picture>
+          <img alt="速度优化秘诀" src="../images/demo6.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1WEcxzFEAT" target="_blank">
        <picture>
-          <img alt="Người ảo số Xiaozhi có đánh thức bằng giọng nói" src="../images/demo8.png" /></picture>
+          <img alt="小智数字人 支持语音唤醒" src="../images/demo8.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1CKVz6UEuB" target="_blank">
        <picture>
-          <img alt="Gọi giữa các thiết bị" src="../images/demo0.png" /></picture>
+          <img alt="设备呼叫设备，打电话" src="../images/demo0.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1C1tCzUEZh" target="_blank">
        <picture>
-          <img alt="Tình huống y tế phức tạp" src="../images/demo1.png" /></picture>
+          <img alt="复杂医疗场景" src="../images/demo1.png" /></picture>
      </a>
    </td>
  </tr>
@@ -87,31 +87,31 @@ Muốn xem hiệu quả sử dụng? Hãy xem video 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1VC96Y5EMH" target="_blank">
        <picture>
-          <img alt="Phát nhạc, tra cứu thời tiết và đọc tin tức" src="../images/demo7.png" /></picture>
+          <img alt="播放音乐查询天气播报新闻" src="../images/demo7.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV12J7WzBEaH" target="_blank">
        <picture>
-          <img alt="Ngắt giữa chừng theo thời gian thực" src="../images/demo10.png" /></picture>
+          <img alt="实时打断" src="../images/demo10.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1Co76z7EvK" target="_blank">
        <picture>
-          <img alt="Nhận dạng vật thể qua ảnh chụp" src="../images/demo12.png" /></picture>
+          <img alt="拍照识物品" src="../images/demo12.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1pNXWYGEx1" target="_blank">
        <picture>
-          <img alt="Bật tắt thiết bị gia dụng" src="../images/demo5.png" /></picture>
+          <img alt="控制家电开关" src="../images/demo5.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1TJ7WzzEo6" target="_blank">
        <picture>
-          <img alt="Nhiệm vụ nhiều lệnh" src="../images/demo11.png" /></picture>
+          <img alt="多指令任务" src="../images/demo11.png" /></picture>
      </a>
    </td>
  </tr>
@@ -119,31 +119,31 @@ Muốn xem hiệu quả sử dụng? Hãy xem video 🎥
    <td>
      <a href="https://www.bilibili.com/video/BV1ZQKUzYExM" target="_blank">
        <picture>
-          <img alt="Điểm truy cập MCP" src="../images/demo13.png" /></picture>
+          <img alt="MCP接入点" src="../images/demo13.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1zUW5zJEkq" target="_blank">
        <picture>
-          <img alt="Gửi lệnh MQTT" src="../images/demo4.png" /></picture>
+          <img alt="MQTT指令下发" src="../images/demo4.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1Exu3zqEDe" target="_blank">
        <picture>
-          <img alt="Nhận dạng vân giọng" src="../images/demo14.png" /></picture>
+          <img alt="声纹识别" src="../images/demo14.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV1CDKWemEU6" target="_blank">
        <picture>
-          <img alt="Giọng nói tùy chỉnh" src="../images/demo2.png" /></picture>
+          <img alt="自定义音色" src="../images/demo2.png" /></picture>
      </a>
    </td>
    <td>
      <a href="https://www.bilibili.com/video/BV12yA2egEaC" target="_blank">
        <picture>
-          <img alt="Trò chuyện bằng tiếng Quảng Đông" src="../images/demo3.png" /></picture>
+          <img alt="使用粤语交流" src="../images/demo3.png" /></picture>
      </a>
    </td>
  </tr>
@@ -170,7 +170,7 @@ Dự án này cung cấp hai phương pháp triển khai, vui lòng chọn theo
 | Phương pháp triển khai | Đặc điểm | Tình huống áp dụng | Tài liệu triển khai | Yêu cầu cấu hình | Video hướng dẫn |
 |---------|------|---------|---------|---------|---------|
 | **Cài đặt tối giản** | Đối thoại thông minh, quản lý đơn tác nhân | Môi trường cấu hình thấp, dữ liệu lưu trong tệp cấu hình, không cần cơ sở dữ liệu | [①Phiên bản Docker](../Deployment.md#%E6%96%B9%E5%BC%8F%E4%B8%80docker%E5%8F%AA%E8%BF%90%E8%A1%8Cserver) / [②Triển khai mã nguồn](../Deployment.md#%E6%96%B9%E5%BC%8F%E4%BA%8C%E6%9C%AC%E5%9C%B0%E6%BA%90%E7%A0%81%E5%8F%AA%E8%BF%90%E8%A1%8Cserver)| `FunASR` cục bộ cần nhiều bộ nhớ; dùng toàn API thì mức dùng tài nguyên thấp | - |
-| **Cài đặt toàn bộ module** | Đối thoại thông minh, quản lý đa người dùng, quản lý đa tác nhân, thao tác qua bảng điều khiển thông minh | Trải nghiệm đầy đủ tính năng, dữ liệu lưu trong cơ sở dữ liệu |[①Phiên bản Docker](../Deployment_all.md#%E6%96%B9%E5%BC%8F%E4%B8%80docker%E8%BF%90%E8%A1%8C%E5%85%A8%E6%A8%A1%E5%9D%97) / [②Triển khai mã nguồn](../Deployment_all.md#%E6%96%B9%E5%BC%8F%E4%BA%8C%E6%9C%AC%E5%9C%B0%E6%BA%90%E7%A0%81%E8%BF%90%E8%A1%8C%E5%85%A8%E6%A8%A1%E5%9D%97) / [③Hướng dẫn tự động cập nhật triển khai mã nguồn](../dev-ops-integration.md) | `FunASR` cục bộ cần nhiều bộ nhớ, thêm tài nguyên cơ sở dữ liệu; dùng toàn API thì mức dùng tài nguyên thấp| [Video hướng dẫn khởi động mã nguồn cục bộ](https://www.bilibili.com/video/BV1wBJhz4Ewe) |
+| **Cài đặt toàn bộ module** | Đối thoại thông minh, quản lý đa người dùng, quản lý đa tác nhân, bảng điều khiển thông minh | Trải nghiệm đầy đủ tính năng, dữ liệu lưu trong cơ sở dữ liệu |[①Phiên bản Docker](../Deployment_all.md#%E6%96%B9%E5%BC%8F%E4%B8%80docker%E8%BF%90%E8%A1%8C%E5%85%A8%E6%A8%A1%E5%9D%97) / [②Triển khai mã nguồn](../Deployment_all.md#%E6%96%B9%E5%BC%8F%E4%BA%8C%E6%9C%AC%E5%9C%B0%E6%BA%90%E7%A0%81%E8%BF%90%E8%A1%8C%E5%85%A8%E6%A8%A1%E5%9D%97) / [③Hướng dẫn tự động cập nhật triển khai mã nguồn](../dev-ops-integration.md) | `FunASR` cục bộ cần nhiều bộ nhớ, thêm tài nguyên cơ sở dữ liệu; dùng toàn API thì mức dùng tài nguyên thấp| [Video hướng dẫn khởi động mã nguồn cục bộ](https://www.bilibili.com/video/BV1wBJhz4Ewe) |

 Câu hỏi thường gặp và hướng dẫn liên quan, vui lòng tham khảo [liên kết này](../FAQ.md)

@@ -193,7 +193,7 @@ Công cụ kiểm tra dịch vụ: https://2662r3426b.vicp.fun/test/
 >
 > 2. `Cấu hình streaming`: Phù hợp với demo, đào tạo, hơn 2 đồng thời, v.v., sử dụng công nghệ xử lý streaming, tốc độ phản hồi nhanh hơn, trải nghiệm tốt hơn.
 >
-> Từ phiên bản `0.5.2`, dự án hỗ trợ cấu hình streaming, so với các phiên bản trước đó, tốc độ phản hồi cải thiện khoảng `2.5 giây`, cải thiện đáng kể trải nghiệm người dùng.
+> Từ phiên bản `0.5.2`, dự án hỗ trợ cấu hình streaming, so với phiên bản đầu, tốc độ phản hồi cải thiện khoảng `2.5 giây`, cải thiện đáng kể trải nghiệm người dùng.

 | Tên module | Cài đặt miễn phí cho người mới | Cấu hình streaming |
 |:---:|:---:|:---:|
@@ -211,7 +211,7 @@ Dự án này cung cấp các công cụ kiểm tra sau để giúp bạn xác m

 | Tên công cụ | Vị trí | Phương pháp sử dụng | Mô tả chức năng |
 |:---:|:---|:---:|:---:|
-| Công cụ kiểm tra tương tác âm thanh | main》digital-human》index.html | Trong `main/digital-human`, cài đặt các phụ thuộc theo README, rồi chạy `python start.py` và mở `http://127.0.0.1:8006/index.html` | Kiểm tra chức năng phát và nhận âm thanh, xác minh xử lý âm thanh phía Python có bình thường không |
+| Công cụ kiểm tra tương tác âm thanh | main》digital-human》index.html | Chạy `python start.py` trong `main/digital-human`, sau đó mở `http://127.0.0.1:8006/index.html` | Kiểm tra chức năng phát và nhận âm thanh, xác minh xử lý âm thanh phía Python có bình thường không |
 | Công cụ kiểm tra phản hồi mô hình | main》xiaozhi-server》performance_tester.py | Thực hiện `python performance_tester.py` | Kiểm tra tốc độ phản hồi của các module cốt lõi như ASR(Nhận dạng giọng nói), LLM(Mô hình lớn), VLLM(Mô hình thị giác), TTS(Tổng hợp giọng nói) |

 > 💡 Gợi ý: Khi kiểm tra tốc độ mô hình, chỉ kiểm tra các mô hình đã cấu hình khóa.
@@ -219,7 +219,7 @@ Dự án này cung cấp các công cụ kiểm tra sau để giúp bạn xác m
 ---
 ## Danh sách tính năng ✨
 ### Đã thực hiện ✅
-![Sơ đồ kiến trúc cài đặt đầy đủ các module](../images/deploy2.png)
+![请参考-全模块安装架构图](../images/deploy2.png)
 | Module chức năng | Mô tả |
 |:---:|:---|
 | Kiến trúc cốt lõi | Dựa trên [cổng MQTT+UDP](https://github.com/xinnan-tech/xiaozhi-esp32-server/blob/main/docs/mqtt-gateway-integration.md), WebSocket, máy chủ HTTP, cung cấp hệ thống quản lý bảng điều khiển và xác thực hoàn chỉnh |
@@ -255,7 +255,7 @@ Xiaozhi là một hệ sinh thái, khi bạn sử dụng sản phẩm này, bạ

 | Phương pháp sử dụng | Nền tảng hỗ trợ | Nền tảng miễn phí |
 |:---:|:---:|:---:|
-| Gọi giao diện openai | Alibaba Bailian, Volcano Engine, DeepSeek, Zhipu, Gemini, iFLYTEK, LM Studio | Zhipu, Gemini |
+| Gọi giao diện openai | Alibaba Bailian, Volcano Engine, DeepSeek, Zhipu, Gemini, iFlytek | Zhipu, Gemini |
 | Gọi giao diện ollama | Ollama | - |
 | Gọi giao diện dify | Dify | - |
 | Gọi giao diện fastgpt | Fastgpt | - |
@@ -281,7 +281,7 @@ Trên thực tế, bất kỳ VLLM nào hỗ trợ gọi giao diện openai đ

 | Phương pháp sử dụng | Nền tảng hỗ trợ | Nền tảng miễn phí |
 |:---:|:---:|:---:|
-| Gọi giao diện | EdgeTTS, iFLYTEK, Volcano Engine, Tencent Cloud, Alibaba Cloud và Bailian, CosyVoiceSiliconflow, TTS302AI, CozeCnTTS, CustomTTS, OpenAITTS, MinimaxTTS | EdgeTTS, CosyVoiceSiliconflow(một phần) |
+| Gọi giao diện | EdgeTTS, iFlytek, Volcano Engine, Tencent Cloud, Alibaba Cloud và Bailian, CosyVoiceSiliconflow, TTS302AI, CozeCnTTS, CustomTTS, OpenAITTS, MinimaxTTS | EdgeTTS, CosyVoiceSiliconflow(một phần) |
 | Dịch vụ cục bộ | FishSpeech, GPT_SOVITS_V2, GPT_SOVITS_V3, IndexStreamTTS, PaddleSpeechTTS | IndexStreamTTS, PaddleSpeechTTS, FishSpeech, GPT_SOVITS_V2, GPT_SOVITS_V3 |

 ---
@@ -299,7 +299,7 @@ Trên thực tế, bất kỳ VLLM nào hỗ trợ gọi giao diện openai đ
 | Phương pháp sử dụng | Nền tảng hỗ trợ | Nền tảng miễn phí |
 |:---:|:---:|:---:|
 | Sử dụng cục bộ | FunASR, SherpaASR, SherpaParaformerASR, VoskASR | FunASR, SherpaASR, SherpaParaformerASR, VoskASR |
-| Gọi giao diện | FunASRServer, Volcano Engine, iFLYTEK, Tencent Cloud, Alibaba Cloud và Bailian, Baidu Cloud, OpenAI ASR, GroqASR, Qwen3ASRFlash | FunASRServer |
+| Gọi giao diện | FunASRServer, Volcano Engine, iFlytek, Tencent Cloud, Alibaba Cloud, Baidu Cloud, OpenAI ASR, GroqASR, Qwen3ASRFlash | FunASRServer |

 ---

@@ -307,9 +307,9 @@ Trên thực tế, bất kỳ VLLM nào hỗ trợ gọi giao diện openai đ

 | Phương pháp sử dụng | Nền tảng hỗ trợ | Nền tảng miễn phí |
 |:---:|:---:|:---:|
-| Gọi giao diện | Dịch vụ nhận dạng vân giọng (voiceprint-api) | - |
+| Gọi giao diện | Dịch vụ nhận dạng giọng nói (voiceprint-api) | - |

-Nhận dạng vân giọng cần dịch vụ [voiceprint-api](https://github.com/xinnan-tech/voiceprint-api) được triển khai riêng (dựa trên 3D-Speaker) và MySQL. Xem [hướng dẫn tích hợp voiceprint](../voiceprint-integration.md).
+Nhận dạng giọng nói theo dấu vân tay cần dịch vụ [voiceprint-api](https://github.com/xinnan-tech/voiceprint-api) được triển khai riêng (dựa trên 3D-Speaker) và MySQL. Xem [hướng dẫn tích hợp voiceprint](../voiceprint-integration.md).

 ---

@@ -355,10 +355,11 @@ Nhận dạng vân giọng cần dịch vụ [voiceprint-api](https://github.com
 | <img src="../images/logo_contributors.png" width="160"> | [Người đóng góp mã](https://github.com/xinnan-tech/xiaozhi-esp32-server/graphs/contributors) | Cảm ơn [tất cả người đóng góp mã](https://github.com/xinnan-tech/xiaozhi-esp32-server/graphs/contributors), sự cống hiến của bạn khiến dự án mạnh mẽ và vững chắc hơn. |


-<a href="https://www.star-history.com/?repos=xinnan-tech%2Fxiaozhi-esp32-server&type=date&legend=top-left">
+<a href="https://star-history.com/#xinnan-tech/xiaozhi-esp32-server&Date">
+
 <picture>
-   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&theme=dark&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
-   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
-   <img alt="Star History Chart" src="https://api.star-history.com/chart?repos=xinnan-tech/xiaozhi-esp32-server&type=date&legend=top-left&sealed_token=cQovHKgZjqEnQ-svfYfN392irNvGuq-6pyv4cA8nd2jQEhQLz1ETV4YHTVk2UZyLMFbCQuZA7jduRh3YbeK5WPYaRLrfmIimGQa3lram652jJL9oQk-UuSZA5H6L4dPIhZc8KCc-Ur_UAUNbly7TePpnTR2otGknBLCOjOliD4fk1st6z7tPEDVjSRx5" />
+   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date&theme=dark" />
+   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date" />
+   <img alt="Star History Chart" src="https://api.star-history.com/svg?repos=xinnan-tech/xiaozhi-esp32-server&type=Date" />
 </picture>
 </a>
@@ -16,6 +16,8 @@ digital-human 是独立的数字人测试模块，负责提供本地测试页面
 pip install -r wakeword_runtime/requirements.txt
 ```

+唤醒词模型要另外下载：仓库里不带模型文件，默认配置（`wakeword_runtime/config.json` 里 `wakeword.enabled` 为 true）下，`wakeword_runtime/models/` 里没有模型文件时启动会打印「警告: 唤醒词服务启动失败」。页面能打开，麦克风的输入流也已经打开，但没有唤醒词检测在消费这些数据，语音唤不醒，页面上的其它功能不受影响。模型下载地址和摆放位置见[唤醒词专题文档](../../docs/digital-human-wakeword.md)。
+
 启动模块：

 ```bash
@@ -31,8 +31,6 @@ class AudioPlugin(Plugin):
            self.source.stop()

    def shutdown(self) -> None:
-        if self.source is not None:
-            self.source.stop()
        if self.app is not None:
            self.app.audio_source = None
        self.source = None
@@ -1,3 +1,6 @@
 sherpa-onnx==1.12.29
 sounddevice>=0.4.4
 pypinyin==0.55.0
+# 唤醒词运行时在 core/detector.py、core/microphone.py 里直接用 numpy，
+# 版本跟 main/xiaozhi-server/requirements.txt 一致，两个模块共用一个环境时不会冲突
+numpy==1.26.4
@@ -51,14 +51,13 @@ class TestRuntimeApplication:
        logger.info("test runtime application started")

    def stop(self) -> None:
-        was_running = self._is_running
-        self._is_running = False
-        if was_running:
-            self.event_bridge.publish_service_stopping()
-            logger.info("test runtime application stopped")
-        # 启动中途失败时 _is_running 还是 False，但插件可能已经开过采集流，
-        # 这里不提前返回，保证采集流一定被关掉
+        if not self._is_running:
+            return
+
+        self.event_bridge.publish_service_stopping()
        self.plugins.stop_all()
+        self._is_running = False
+        logger.info("test runtime application stopped")

    def shutdown(self) -> None:
        self.stop()
@@ -2,7 +2,7 @@ import os
 import importlib.util
 import asyncio

-print("使用前请根据仓库根目录下 docs/performance_tester.md 的说明准备配置。")
+print("使用前请根据docs/performance_tester.md的说明准备配置。")


 def list_performance_tester_modules():
Author	SHA1	Message	Date
hrz	aa2f44dac6	唤醒词文档里 tokens.txt 的行数改成 227 实际包里的 tokens.txt 是 227 行，之前写的 228 行不对，按实测值改过来。	2026-10-27 10:23:37 +08:00
hrz	c2c050dbd0	唤醒词文档改准：int8 的定位和 decoder 的版本 int8 只决定选方案一还是方案二，不是 ModelScope 下载的前提，运行时只按四个固定文件名找模型。官方包里 decoder 也有 int8 版，不是只有 ModelScope 那份才有，方案二里 decoder 仍用 fp32 是因为量化对 decoder 收益不大。	2026-10-27 09:13:42 +08:00
hrz	04f8b17d97	文档写清 ModelScope 换模型要连 avg 一起换方法2 下载的仓库里只有 epoch-12 的模型，文件名里是 avg-2 不是 avg-1。原来只写把 epoch-99 换成 epoch-12，照着换会得到不存在的文件名，cp 直接报错。现在写明要换成 epoch-12-avg-2，并说明仓库里缺 int8 那两份时方法2 不能用，只能改用方法1。	2026-10-26 22:12:21 +08:00
hrz	c822648882	唤醒词文档补上拷贝改名这步并改对精度说法 ModelScope 下的文件留在自己的目录里，运行时只认 models/ 下的四个固定名字，不拷进去改名就会报找不到模型。官方包里 decoder 只有 fp32 一版，速度优先那套写成了 int8 三件套，跟同文件的注意事项打架，一并改对。	2026-10-26 21:38:47 +08:00
hrz	4ec37c6e33	文档改准：缺模型时是唤醒词没起来，不是麦克风坏了评审指出原来那两句说反了。音频插件先把麦克风打开，唤醒词插件才去找模型文件，缺文件的是后者，所以是检测没起来、数据没人消费，麦克风一直在采。keywords.txt 也是启动时自己生成，不用提前准备。	2026-10-26 21:32:37 +08:00
hrz	7af5630b2a	在数字人文档里写清唤醒词模型的前置条件模型文件不在仓库里，默认配置下缺模型只会打印一句唤醒词服务启动失败，页面能开但麦克风采集起不来。digital-human 的 README、唤醒词专题文档和 paddlespeech 部署文档三处都补上这个前置条件，并指向模型下载那一节。	2026-10-26 21:16:47 +08:00
hrz	43f30c06c1	给唤醒词运行时补上 numpy 依赖 core/detector.py 和 core/microphone.py 在模块层直接 import numpy， requirements.txt 里没有这一条。干净环境里照文档装完依赖就启动，会报 ModuleNotFoundError，页面都开不起来。版本跟 xiaozhi-server 那份一样，两个模块共用一个环境时不会冲突。	2026-10-26 21:16:47 +08:00