In the last couple of weeks (presumably driven by the release of Deepseek), a number of low cost ESP32 voice assistants have arrived, eg.,
https://www.aliexpress.com/item/1005008600891141.html
https://www.aliexpress.com/item/1005008634826817.html
https://www.aliexpress.com/item/1005008634051137.html
Has anyon e tried any of those and any comments on how well they work or even how far we are from using them with ESPHome and microwakeword?
Interesting ā¦ looks good ā¦ based on ESP32-S3 (like the Voice Assist PE) so in theory shouldnāt be too hard to run ESPHome.
Interesting. Iāve bought one. Will have a play and report back.
Which one did you get?
Iāll be very interested to hear, sadly currently lack the time to go and hack at themā¦
The one in the first link. Thought it looked the best. Havenāt received it yet though.
OK. So I got this device: DeepSeek XiaoZhi AI Voice Chat Robot ESP32-S3 1.28 inch LCD N16R8 Development Board Astronaut Clock Desktop Ornament - AliExpress 7
It appears that the documentation for this device is available here: Moji å°ęŗAIč”ēē - ē«åå¼ęŗē”¬ä»¶å¹³å°
This is very useful as it provides all of the pinouts etc. Edit: See my later post for the correct information. Iām afraid that what I posted above was for a similar but different device, which resulted in wasted time working with the wrong pinouts.
I have successfully installed ESPHome on the device. All the standard stuff works, like wifi, ota etc, but in terms of device specific hardware I have thus far only managed to get the backlight to work. The display is listed as a GC9A01 display, but none of the various permutations I have tried have yielded any results. If anyone knows which ESPHome component I should be using with this display I would be most grateful. In the meantime, Iāll continue with my trial and error approach. Is there a better way? Advice from googling and asking Copilot has thus far proved incorrect.
Unfortunately the devices sold on AliExpress are inexpensive clones that just use a MAX98357, INMP441, GC9A01 and a single WS2812.
You can find more details of the product here: DeepSeek XiaoZhi AI Voice Chat Robot ESP32-S3 1.28 inch LCD N16R8 Development Board Astronaut Clock Desktop Ornament
I have successfully got the ESPHome Voice PE template and the S3-BOX-3 template running with a few minor tweaks.
Hi. Thanks for that. Iāve independently found the same information, after quite a lot of digging. One of the problems is that there are a lot of versions of this kicking around, and it is quite hard to find the correct one. Iāve found the source code accessible here and Iāve successfully compiled it and downloaded it to the device, so I can confirm it is the correct one for that device. It also took me a little while to figure out which of the many boards listed in the source code was the correct one: answer ābread-compact-wifiā. Having names in Chinese makes things entertaining! Anyway, once I figured out the board type, that gave me the pin assignments and I was able to make quite a lot of progress mapping this onto an ESPHome project. I have the display working fine, and also the speaker. (Though curiously first this worked in Music Assistant but not Home Assistant, and now it is vice versa! Still tracking that one down). At the moment my microphone doesnāt seem to work, but Iāll follow up your tip on using the s3-box-3 template and see whether that helps.
While I had the device on the stock firmware, I couldnāt get the wakeword functionality to work. It only started listening when I pressed the boot button at the back. I suspect this is all Iāll be able to do on ESPHome as well.
If you flip your device over, does it have 3 screws or 1 singular? Does it have a sticker that says v4?
The schematic shows the pinouts for the components, once I mapped the esphome templates to the correct GPIO it ājust workedā you should be able to get wakeword working, but I found the microphone position inside the case makes struggle to listen unless youāre right next to it
Yes, it has 3 screws and a sticker that says V4. Which template are you using? Is this this one?
You will need to remove the i2c, audio_adc/es7210, audio_dac/es8311 and switch the images to use https://github.com/jptrsn/wake-word-voice-assistants/tree/main/casita
You will need to remove the voice_kit, audio_dac, external_components. This will have no screen output, but work similarly to the official device
I have based mine template on voice pe with the display bits from the box s3 ported across, so ideally you will want to merge bits once tested independently that theyre working.
I have a āV2ā versionā¦
Could you share your ESPHome config?
@lareeth I just received the XiaoZhi āV5 ENā device. I got most of it working but audio is still missing. How did you manage to interface the audio ic? Is it using the ES8311?
V5 is v1.62 of the software, nothing changed with hardware, and the pins are the same for all versions of this cube and some other models too like this one:
https://www.aliexpress.com/item/1005008720707304.html
except ofcourse, that one has no display.
also same pins as they also use for their standard breadboard version.
So, I opened up the device ā hardware looks identical.
Itās still using the MAX98357A instead of the ES8311, just like yours.
But for some reason, I still canāt get any audio output working. Any idea whats wrong?
Hereās my code:
i2s_audio:
- id: i2s # For microphone
i2s_lrclk_pin: GPIO4 #WS
i2s_bclk_pin: GPIO5 #SCK
- id: speaker_i2s # For speaker
i2s_lrclk_pin: GPIO16
i2s_bclk_pin: GPIO15
microphone:
- platform: i2s_audio
i2s_audio_id: i2s
adc_type: external
i2s_din_pin: GPIO6 #SD
id: va_mic
channel: left
pdm: false
bits_per_sample: 16bit
speaker:
platform: i2s_audio
id: va_speaker
i2s_audio_id: i2s
dac_type: external
i2s_dout_pin:
number: GPIO7 # DIN Pin of the MAX98357A Audio Amplifier
channel: mono
Same hardware and circuit diagram,
This configuration can produce sound normally in my environment
esp32:
board: esp32-s3-devkitc-1
framework:
type: esp-idf
i2s_audio:
- id: i2s # For microphone
i2s_lrclk_pin: GPIO4 #WS
i2s_bclk_pin: GPIO5 #SCK
- id: speaker_i2s # For speaker
i2s_lrclk_pin: GPIO16
i2s_bclk_pin: GPIO15
microphone:
- platform: i2s_audio
i2s_audio_id: i2s
adc_type: external
i2s_din_pin: GPIO6 #SD
id: va_mic
speaker:
platform: i2s_audio
id: va_speaker
i2s_audio_id: i2s
dac_type: external
i2s_dout_pin:
number: GPIO7 # DIN Pin of the MAX98357A Audio Amplifier
media_player:
- platform: speaker
name: None
announcement_pipeline:
speaker: va_speaker
format: WAV
volume_min: 0.4
Hello everyone,
Iām thinking on replacing my alexaās without breaking the bank, do you think this is a good solution?
Howās the sound and how does the microphone catches wakeup words?
Hope you can help me out.
i will add the display part to your code:
time:
- platform: homeassistant
id: homeassistant_time
spi:
clk_pin: GPIO14
mosi_pin: GPIO17
interface: hardware
id: spihwd
font:
- file: "gfonts://Roboto"
id: my_font
size: 28
- file: "gfonts://Roboto"
id: roboto_52
size: 52
image:
- file: images/ow480x480.jpg
id: my_image
resize: 240x240
type: RGB
display:
- platform: ili9xxx
invert_colors: true
model: GC9A01A
rotation: 0
reset_pin: GPIO18 # (Pin on display - RESET)
cs_pin: GPIO13 # (Pin on display - CS)
dc_pin: GPIO10 # (Pin on display - DC)
lambda: |-
auto red = Color(255,0,0), green = Color(0,255,0), blue = Color(0,0,255), yellow = Color(255,255,0), white = Color(255,255, 255), cyberlightgreen = Color(0,255,159), cyberlightblue = Color(0,184,255), cyberpink = Color(214,0,255);
it.image(0, 0, id(my_image));
it.strftime(120, 120, id(roboto_52), cyberlightgreen, TextAlign::CENTER, "%H:%M:%S", id(homeassistant_time).now());
this will show a image with the time on top.
This is V5-EN version?
itās a V4 updated to V5 since the version is the software, not hardware. v4 was 1.60, v5 is 1.62 (current) of the xiaozhi code, but since the one in that image is flashed ith esphome itās none of them, if anything itās the version of esphome 2025.5
