Smart speakers are no strangers to everyone. Nowadays, there are too many such as Tmall Genie, Xiaodu, and Xiaoai. However, currently these smart speakers are really stupid. They either answer irrelevantly or simply pretend to be dead and ignore you.
In the past two years, AI has been extremely popular. So, some people have connected AI large models to smart speakers to achieve more intelligent communication. Thus, this year’s project – MIGPTserver was born. The MIGPT on the project page is a very mature project. It has functions such as AI question answering, role-playing, streaming response, short-term and long-term memory, and customizable TTS.
Below is the selection of smart speakers. In some categories, some speakers cannot run, and some speakers do not support continuous conversation. This point still needs attention. For example, the enhanced version of Xiaoai Classmate used at Panda’s home does not support continuous conversation. After choosing a good speaker, fill in your Xiaomi account ID, password, and corresponding device ID below. The commands below do not need to be changed. After choosing a good speaker, it has been set up for you by default. Then there are the settings for word conversation and continuous conversation. Since my speaker does not support continuous conversation, Panda directly turns this off here. For word conversation, you can set up invocation words. However, the use of invocation words requires you to wake up Xiaoai first and then say it. For example, if you want to ask Xiaoai about today’s weather, you need to ask: ‘Xiaoai Classmate, may I ask what the weather is like today?’ At this time, Xiaoai will answer you through the large model instead of its own intelligence. Finally, there are the settings for AI service and TTS engine. In the AI model section, OpenAI, Zhipu, Doubao, Qianwen, deepseek, Zero One Everything, Baichuan, and kimi are supported. Each model provides a default docking address, and the secret key and model need to be set by yourself. However, the author has very carefully integrated how to obtain the model into the control panel. TTS is a voice service that supports dege reading, Doubao, and OpenAI. After setting everything up, click Start. At this time, asking Xiaoai Classmate according to your settings can get answers from the large model. These can also be seen in the Docker logs. The deployment process of migpt-server is not difficult. Before deployment, let’s open Docker, click on Images and then switch to the Repository interface. You can see Settings in the upper right corner. Opening Settings can show Repository Settings. Here we can set up image acceleration. For repository settings, some good acceleration sources are recommended, such as docker. 1panel.live, dockerpull, jijiai, etc. Click Add, enter our repository address, and then confirm. It is added. It should be noted that after adding, remember to check the acceleration image source that needs to be used, otherwise it will not take effect. After preparation, enter lmk123/migpt-server here for custom pulling. After downloading, switch to the local repository and you can see the image. Double-click the just-downloaded image. Only need to map the port to the local area. Do not change the container port.
Finally, start the container after port mapping. Enter the IP address of ZSpace followed by ‘:36592’ in the browser to access the control panel.
After debugging your Xiao Ai, you can call the AI large model. In conclusion, this is a very good project. The previously less capable Xiao Ai can finally know everything. However, continuous conversation is not supported here for Panda, which is quite troublesome. If there is a need, you can find a speaker that supports continuous conversation for a more comfortable experience.
