This repo contains custom scripts and patches, to make a custom rootfs image free of propietary software, and installing open source programs.
Patches and scripts can be tweaked, so if really want, you can keep the original functions of the Chinese XiaoAi assistant, and install new programs as well.
The main purpose of this toolset is to have your DIY Voice Assistant (Alexa-like) and Smart Speaker, with lots of integrations for music and automation.
By using MPD, Snapcast, Shairport-Sync, Upmpdcli you can make your speaker a full media player compatible with multiple cast protocols,
and also have a voice assistant powered by Porcupine and Whisper that can interact with your Home Assistant.
Everything powered by Open Source Software!
| Model | Name | Target version (recommended) |
|---|---|---|
 |
Xiaoai Speaker Pro (black, infrared) | 1.74.10 |
 |
Xiaoai Speaker (white, no infrared) | 1.74.10 (build as LX06) |
 |
Xiaomi Mi AI Speaker Mini | 1.32.6 |
 |
Xiaoai Speaker Art (China, white) | 1.76.4 |
 |
Xiaoai Speaker Art Battery (Black) | |
 |
Xiaomi Mi Smart Speaker (Global, Google Assistant) | |
 |
Xiaoai Speaker Play | |
 |
Xiaoai Speaker Play Enhanced Edition, no Clock LED (NuttX) - details | |
 |
Xiaoai Speaker Play Enhanced Edition, with Clock LED (NuttX) | |
 |
Xiaomi Smart Speaker IR Control (Global, Google Assistant) | |
 |
Xiaomi Mi AI Speaker 2 Gen | |
 |
Xiaomi Sound by HARMAN AudioEFX | |
 |
Redmi XiaoAI Speaker Play | |
 |
Xiaomi Sound Move | |
 |
Xiaomi Mi AI Speaker 1 Gen (?) - S12 | |
 |
Xiaomi XiaoAI HD | |
 |
Yeelight Voice Control Assistant |
Tip
Target version is the recommended version to ensure all patches are working,
but other versions may be still supported and working.
🔧 This is still Work in Progress.
Some new speakers or firmware upgrades change the rootfs partition and include a DER certificate to verify the system.
This may block any changes on non-signed squashfs. Recommended to NOT flash, you may have an invalid rootfs and potentially lock yourself!
You can check this by running binwalk if it contains a Certificate entry:
DECIMAL HEXADECIMAL DESCRIPTION
--------------------------------------------------------------------------------
0 0x0 Squashfs filesystem, little endian, version 4.0, compression:xz, size: 32240378 bytes,
2430 inodes, blocksize: 262144 bytes, created: 2021-04-28 06:34:34
32243716 0x1EC0004 Certificate in DER format (x509 v3), header length: 4, sequence length: 830
Warning
Supported and Tested setup is Linux OS with amd64 arch.
Other hosts (Apple M1, Windows) are not tested nor supported.
Feel free to make a PR if you have a fix for this.
Important
If using WSL, be sure to enable setCaseSensitiveInfo.
You will also need the following tools:
squashfs-tools, providesunsquashfsandmksquashfsmtd-utils(Only for model S12), provides utils for UBI imagesmakersyncshasum- Docker
- patience - package build takes more than 1 hour. :)
Get a copy of your rootfs filesystem from your speaker. It can also be from a system upgrade file.
It should be something similar as this:
nc -vlp 8888 > $HOME/backup-image
# -----
dd if=/dev/mtd4 of=/tmp/image
nc $IP_ADDR 8888 < /tmp/imageSince the speakers filesystem is read-only format, we have to reflash it to add the new applications and patches. There are three steps to perform this: extract, patch, build.
Optionally but recommended, prepare the packages you want to install by editing the packages.sh script.
Build the docker image and run it to build all the packages. Probably it will take more than an hour.
Caution
Run the build packages.sh process with Docker, since the package build performs some patching to the system, otherwise it could harm your GNU/Linux installation.
docker build -t xiaoai-patch packages
docker run -it -v $PWD:/xiaoai xiaoai-patchYou can now run the commands to prepare the new image.
sudo make clean
sudo make extract FILE=image-mtd4
sudo make patch MODEL=lx06
sudo make build MODEL=lx06
# or all at once :)
sudo make clean all FILE=image-mtd4 MODEL=lx06Tip
Ensure the image format is correct, by comparing the original and new images. Use file or other commands to check info.
After you have the new image ready, send it to the speaker, and flash the not-in-use rootfs partition, boot it and test.
You should have some wires soldered to the board to perform TTL in case it is required.
As long as you perform steps as described and not flashing content in wrong partitions, you can reverse failed boot with Uboot safely.
In order to enable Uboot menu, check in the environment partition that you have the setting bootdelay=3.
If is set to bootdelay=0 then Uboot will continue normal boot process and you won't be able to stop it unless you get into fastboot or recovery mode. (?)
In most cases, binary program is fw_setenv or fw_env and data is set into first partition as string.
Ensure you can access Uboot before writing changes, it is your rescue!
Hit any key to stop autoboot: 0
YOU are responsible for any use or damage this software may cause. This repo and its content is intended for educational purposes only. Use at your own risk.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent
