Xiaozhi ESP-32 Back-end Service (xiaozhi-esp32-server)

（中文 | English）

This project provides backend services for the open-source smart hardware project xiaozhi-esp32。Implemented in Python following theXiaozhi Communication Protocol。

Target Audience

This project requires compatible esp32 hardware devices. If you have purchased esp32 hardware, successfully connected to Xiage's deployed backend, and wish to independently set up the xiaozhi-esp32 backend service, this project is for you.

To see a demo, watch this video:

To fully experience this project, follow these steps:

Prepare hardware compatible with the xiaozhi-esp32 project. For supported models, click here.
Use a computer/server with at least 4-core CPU and 8GB RAM to run this project. After deployment, you'll see the service endpoint address in the console.
Download the xiaozhi-esp32 project, replace the default endpoint address with your own, compile, and flash the firmware to your device.
Start the device and check your server console logs to verify successful connection.

Warning

This project has been established for a short time and has not passed the network security assessment, so please do not use it in the production environment.

Feature List

Implemented

xiaozhi-esp32 WebSocket communication protocol
Supports wake-word initiated dialogue, manual dialogue, and real-time interruption of dialogue.
Support for 5 languages: Mandarin, Cantonese, English, Japanese, Korean (FunASR - default)
Flexible LLM switching (openai:ChatGLM - default, Aliyun, DeepSeek; dify:Dify)
Flexible TTS switching (EdgeTTS - default, ByteDance Doubao TTS)

In Progress

Sleep mode after inactivity
Dialogue memory
Change the mood mode

Supported Services

Type	Service	Usage	Pricing Model	Notes
LLM	Aliyun	openai API call	Token-based	Apply for API Key
LLM	DeepSeek	openai API call	Token-based	Apply for API Key
LLM	Bigmodel	openai API call	Free	Create API Key
LLM	Dify	dify API call	Token-based	Self-hosted
TTS	HuoshanTTS	API call	Token-based	Create API Key
TTS	EdgeTTS	API call	Free
VAD	SileroVAD	Local	Free
ASR	FunASR	Local	Free

In fact, any LLM that supports OpenAI API calls can be integrated and used.

Deployment

This project supports rapid deployment of docker and local source code operation. If you want to have a quick experience, it is recommended to use docker to deploy. If you want to have an in-depth understanding of this project, it is recommended to run the local source code.

Method 1: Quick deployment of docker

The docker image has supported the CPU of x86 architecture and arm64 architecture, and supports running on Chinese operating systems.

Install docker

If your computer has not installed docker, you can follow the tutorial here to install it:Install docker

Create a directory

After installation, you need to find a directory for the configuration file for this project. Let's call it the project directory for the time being. This directory is preferably a newly created empty directory.

Download the configuration file

Open with a browserThis link。

On the right side of the page, find the button named RAW, next to the RAW button, find the download icon, click the Download button, and download the config.yaml file. Download the file to your project directory.

Configure Project

Modify the config.yaml file to configure the various parameters required for this project. The default LLM uses ChatGLMLLM, you need to configure the key to start. The default TTS uses EdgeTTS. This does not require configuration. If you need to replace it withDoubao TTS, you need to configure the key.

Configuration description: This is the default component of each function, such as LLM default to use the ChatGLMLLM model. If you need to switch the model, it is the corresponding name.

The default configuration of this project is only the lowest operating cost configuration（glm-4-flashandEdgeTTSare free），If you need to be better and faster, you need to combine the use of the deployment environment to switch the use of each component。

selected_module:
  ASR: FunASR
  VAD: SileroVAD
  LLM: ChatGLMLLM
  TTS: EdgeTTS

For example, to modify the components used by the LLM, it depends on which LLM API interfaces are supported by this project. Currently, the supported ones are openai and dify. We welcome validation and support for more LLM platforms' interfaces. When using it, change the selected_module to the corresponding name of the following LLM configurations:

LLM:
  AliLLM:
    type: openai
    ...
  DeepSeekLLM:
    type: openai
    ...
  ChatGLMLLM:
    type: openai
    ...
  DifyLLM:
    type: openai
    ...

Some services, for example, if you use the TTSof thedify and bean bags, you need a key, remember to add the configuration file!

Execute the docker command

Open the command line tool, cd enter your project directory, and execute the following command

#If you are Linux, execute
ls
#If you are Windows, execute
dir

If you can see the config.yaml file, you have indeed entered the project directory, and then execute the following command:

docker run -d --name xiaozhi-esp32-server --restart always --security-opt seccomp:unconfined -p 8000:8000 -v $(pwd)/config.yaml:/opt/xiaozhi-esp32-server/config.yaml ccr.ccs.tencentyun.com/xinnan/xiaozhi-esp32-server:latest

If executed for the first time, it may take several minutes, and you have to be patient to wait for it to complete the pull. After normal pulling is completed, you can execute the following command on the command line to see if the service is started successfully.

docker ps

If you can see xiaozhi-server, it means that the service starts successfully. Then you can further execute the following command to view the service log

docker logs -f xiaozhi-esp32-server

If you can see, similar to the following logs, it is a sign that the service of this project is successfully launched.

2025-xx-xx xx:51:59,492 - core.server - INFO - Server is running at ws://xx.xx.xx.xxx:8000
2025-xx-xx xx:51:59,516 - websockets.server - INFO - server listening on 0.0.0.0:8000

Next, you can start compiling esp32 firmware. Please go down and turn to the relevant chapter on compiling esp32 firmware. So since you are deploying with docker, you have to check the IP of your native computer by yourself. Normally, assuming your ip is 192.168.1.25, then your interface address is: ws://192.168.1.25:8000. This information is very useful, and it is required to compile esp32 firmware later.

Method 2 : Local Source Code Deployment

1.Install Prerequisites

This project uses 'conda' to manage dependencies, and after installation, start executing the following commands:

conda remove -n xiaozhi-esp32-server --all -y
conda create -n xiaozhi-esp32-server python=3.10 -y
conda activate xiaozhi-esp32-server

After executing the above command, if your computer is Windows or Mac, execute the following statement:

conda activate xiaozhi-esp32-server
conda install conda-forge::libopus
conda install conda-forge::ffmpeg

If your computer is ubuntu, execute the following statement:

apt-get install libopus0 ffmpeg

2.Install Dependencies

# Clone the project
cd xiaozhi-esp32-server
conda activate xiaozhi-esp32-server
pip config set global.index-url https://mirrors.aliyun.com/pypi/simple/
pip install -r requirements.txt

3.Download ASR Model

Download SenseVoiceSmall to model/SenseVoiceSmall.

By default, the SenseVoiceSmall model is used to convert voice to text. Because the model is large, it needs to be downloaded independently. After downloading, place the model.pt file in the model/SenseVoiceSmall directory. Choose any of the following two download routes.

Line 1: Download Ali Magic TowerSenseVoiceSmall
Line 2: Baidu Netdisk downloadSenseVoiceSmall 提取码: qvna

4.Configure Project

Modify the config.yaml file to configure the various parameters required for this project. The default LLM uses ChatGLMLLM, you need to configure the key to start. The default TTS uses EdgeTTS. This does not require configuration. If you need to replace it withDoubao TTS, you need to configure the key.

Configuration description: This is the default component of each function, such as LLM default to use the ChatGLMLLM model. If you need to switch the model, it is the corresponding name.

The default configuration of this project is only the lowest operating cost configuration（glm-4-flashandEdgeTTSare free），If you need to be better and faster, you need to combine the use of the deployment environment to switch the use of each component。

selected_module:
  ASR: FunASR
  VAD: SileroVAD
  LLM: ChatGLMLLM
  TTS: EdgeTTS

For example, to modify the components used by the LLM, it depends on which LLM API interfaces are supported by this project. Currently, the supported ones are openai and dify. We welcome validation and support for more LLM platforms' interfaces. When using it, change the selected_module to the corresponding name of the following LLM configurations:

LLM:
  AliLLM:
    type: openai
    ...
  DeepSeekLLM:
    type: openai
    ...
  ChatGLMLLM:
    type: openai
    ...
  DifyLLM:
    type: openai
    ...

Some services, for example, if you use the TTSof thedify and bean bags, you need a key, remember to add the configuration file!

5.Run the Project

Run the Project

# Make sure to execute in the root directory of this project
conda activate xiaozhi-esp32-server
python app.py

You'll see the WebSocket endpoint in logs:

2025-xx-xx xx:51:59,492 - core.server - INFO - Server is running at ws://192.168.1.25:8000
2025-xx-xx xx:51:59,516 - websockets.server - INFO - server listening on 0.0.0.0:8000

Among them, the ws://192.168.1.25:8000 is the interface address provided by this project. Of course, your own machine is different from mine. Remember to find your own address.

Compile ESP32 Firmware

Download xiaozhi-esp32 project, configure the project environment according to this tutorial " Windows builds ESP IDF 5.3.2 Development Environment and Compiles Xiaozhi " Cure
Open the xiaozhi-esp32/main/kconfig.projbuild file, find the content of the websocket_url default, change the wss: // api.tenclass.net to your own address, such as

Before modification:

config WEBSOCKET_URL
    depends on CONNECTION_TYPE_WEBSOCKET
    string "Websocket URL"
    default "wss://api.tenclass.net/xiaozhi/v1/"
    help
        Communication with the server through websocket after wake up.

After modification (example):

config WEBSOCKET_URL
    depends on CONNECTION_TYPE_WEBSOCKET
    string "Websocket URL"
    default "ws://192.168.1.25:8000/xiaozhi/v1/"
    help
        Communication with the server through websocket after wake up.

Configure build settings:

# The terminal command line enters the root directory of xiaozhi-esp32
cd xiaozhi-esp32
# For example, the board I use is ESP32S3, so the compile target is ESP32S3. If your board is other models, please replace it with the corresponding model
idf.py set-target esp32s3
# Enter the menu configuration
idf.py menuconfig

After entering the menu configuration, then enter xiaozhi assistant, set the connection_type to websocket Go back to the main menu, then enter xiaozhi assistant, set the BOARD_TYPE of your board Save exit and return to the terminal command line.

Build and package:

idf.py build
cd scripts
python release.py

After the compilation is successful, the firmware file merged-binary.bin is generated in the build directory in the project root directory. This merged-binary.bin is the firmware file that will be recorded on the hardware.

Flash Connect the ESP32 device to the computer, use the Chrome browser, and open the following URL

https://espressif.github.io/esp-launchpad/

Open this tutorial, Flash Tools/Web -side Burning Folding Step (No IDF Development Environment). Turn to: Method 2: ESP-LAUNCHPAD browser web-end burning, start from 3. Burning firmware/download to the development board, follow the tutorial operation.

FAQ

1、TTS often fails, often overtime

Suggestion: If the Edgetts is slow or often fails, you can replace it with a bean bag TTS` with a volcanic engine. If both are slow, the network environment may need to be optimized.

2、Big model reply is a bit slow

Suggestions: Both big models and TTS are dependent interfaces. If the network environment is not good, you can consider changing the local model. Or try to switch different interface models.

3、Why is my ChatGLMLLM replying to a bit? Obviously it is Xiaozhi, but treats me as Xiaozhi.

Suggestion: You can modify the prompts in the configuration file first. You can also replace the free GLM-4-FLASH to the model of other toll versions of ChatGlm.

4、I want to control the operation of electric lights, air conditioners, remote switching and other operations through Xiaozhi.

Suggestion: In the configuration file, set the LLM toDifyLLM, and then arrange the smart application by the Dify.

5、I said very slowly, I paused, Xiaozhi always grabbed me, what to do.

Suggestion: In the configuration file, find this section, change the min_silence_duration_ms value, such as change to 1000.

VAD:
  SileroVAD:
    threshold: 0.5
    model_dir: models/snakers4_silero-vad
    min_silence_duration_ms: 700  # 如果说话停顿比较长，可以把这个值设置大一些

6、For more questions, contact us to feedback

Acknowledgments

This project is inspired by the Bailin Voice Dialogue Robot project, and the basic idea of the project is completed。
Thanks to [Tencent Cloud] (https://cloud.tencent.com/) for providing free docker space for this project。
Thanks to tenclassProvide adequate documentation support on Xiaozhi Communication Protocol。

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

README_en.md

README_en.md

Xiaozhi ESP-32 Back-end Service (xiaozhi-esp32-server)

Target Audience

Warning

Feature List

Implemented

In Progress

Supported Services

Deployment

Method 1: Quick deployment of docker

Method 2 : Local Source Code Deployment

1.Install Prerequisites

2.Install Dependencies

3.Download ASR Model

4.Configure Project

5.Run the Project

Compile ESP32 Firmware

FAQ

1、TTS often fails, often overtime

2、Big model reply is a bit slow

3、Why is my ChatGLMLLM replying to a bit? Obviously it is Xiaozhi, but treats me as Xiaozhi.

4、I want to control the operation of electric lights, air conditioners, remote switching and other operations through Xiaozhi.

5、I said very slowly, I paused, Xiaozhi always grabbed me, what to do.

6、For more questions, contact us to feedback

Acknowledgments

Files

README_en.md

Latest commit

History

README_en.md

File metadata and controls

Xiaozhi ESP-32 Back-end Service (xiaozhi-esp32-server)

Target Audience

Warning

Feature List

Implemented

In Progress

Supported Services

Deployment

Method 1: Quick deployment of docker

Method 2 : Local Source Code Deployment

1.Install Prerequisites

2.Install Dependencies

3.Download ASR Model

4.Configure Project

5.Run the Project

Compile ESP32 Firmware

FAQ

1、TTS often fails, often overtime

2、Big model reply is a bit slow

3、Why is my ChatGLMLLM replying to a bit? Obviously it is Xiaozhi, but treats me as Xiaozhi.

4、I want to control the operation of electric lights, air conditioners, remote switching and other operations through Xiaozhi.

5、I said very slowly, I paused, Xiaozhi always grabbed me, what to do.

6、For more questions, contact us to feedback

Acknowledgments