made a little card for voice assist for these,
1 Like
Hi guys, I have version 3, I flashed it to version 2 but it doesn’t hear me. Any ideas?
could be because xiaozhi have a amazing pipeline, really fast and with features we don’t have in esphome, like person recognition and SER (speech emotion recognition) and is VERY fast making human like conersations more natural. AND perfect for those that do NOT want to run any AI themself.
another bonus is you can bring your xiaozhi device with you and still control your home without using any VPN 
2 Likes
I’m using the SpotPear ESP32-S3-N16R8 with xiaozhi firmware v1.6.2.
I’m facing an issue where the device enters a sleep mode? (red LED on, screen off) and the default wake word “hi esp” doesn’t work. In normal standby (screen on), the wake word functions correctly. Any suggestions?
@liminal MCP via xiaozhi works perfectly for me in German. I think it gives you a much better chance to control devices using natural spoken language compared to the fixed standard sentences in native ESPHome.
1.62 is quite old firmware (1.8.5 now), you should look into compiling your own from source, then you can also set the wakeword to something better then “hi esp” which it is horrible at catching. “Jarvis” works so much better for me.
here’s a short description on how to install latest xiaozhi firmware using esp-idf:
1 Like
Thanks for your advice! I followed your recommendation and updated to firmware 1.8.5 by compiling from source using ESP-IDF. I also changed the wake word to “computer” instead of the default “hi esp.”
Unfortunately, the recognition isn’t as good as I hoped—it still misses the wake word quite often. I might try switching to “Jarvis”, as you suggested, to see if it performs better.
Is there any difference between releases? I used original 78/xiaozhi github repo?
By the way, the sleep mode issue still persists in 1.8.5.
Greetings
Hey guys, have you also problems with the mic? I bought Ball V2 and it is creaking, whisper doesnt understand what i’m talking. I tried settings with noice reduction but it doesnt helped. Has someone ideas for good cheap alternative mic?
I’m also using a SpotSpear Ball V2. Where can I find instructions to compile the source myself? It seems there’s no precompiled firmware image available for this hardware — or am I mistaken?
Thanks !
Here, are instructions. Its pretty simple scroll to Xiaozhi Compilation.
Greetings and good luck 
1 Like
I’ve compiled the 1.8.8 version and pushed it to the ball, has anyone found instructions how to get back to the AP mode where you can select the wifi?
Can you please explain how did you fix it? For me it doesn’t work when HA https is enabled with letsencrypt certificates.
Hi
This yaml works for you on ESPhome?
Hi
This yaml works for you on ESPhome?
I have the same problem. Voice recordings through the phone app sound perfect. SpotBall v2 produces a lot of interference. Only after installing the model in wyoming-faster-whisper on small-int8 did it start recognizing much better. I haven’t figured out how to get rid of the noise in the microphone yet. Do you experience clicking sounds when you activate the speaker or when the device detects the activation phrase?
My devices do this when using an under or over powered adaptor.
@RealDeco I am having so much fun on your webpage. I was trying to get AI to write code and it was hours and hours of hallucinating. I am going to try out your Harley personality next and maybe pickup a different device, is the ball v2 the best cheap satellite device available? Also guys, he has code on his page so you can change the Ai personality with a voice request. This way you can go beyond the 2 choice limit with wake words that ha offers.
1 Like
I’d like to share my modifications of the original Ball_v2.yaml from @RealDeco
What it changes:
1 - Adds controls for auto dim (turn on/off) the screen. It lights up upon wake word and dims after N seconds to configurable brightness settings
2 - Adds controls for wake light, using the embeded RGB Led (but separate settings) so it lights up on wake word, fast blinks on “thinking” and slow blinks during response. It just mimics the echo devices behavior for a more polished interaction. The embedded LED goes back to it’s previous setting after the interaction
3 - Least but not last, it integrates the media_player component for future uses as a output speaker for Music Assistant for example. The config file is fully functional, but the full duplex/ media stream DO NOT WORK currently. I’m working with @meconiotech in his absolute amazing full duplex i2s audio driver, so even simple devices like the spotpears in wich mic and speaker share a single i2s bus can have full duplex audio (and the wake word can be always on even with music playing or to interrupt responses). In the future, using his intercom project even drop-in style intercom should be possible.
The file can be found here:
1 Like
nice!, i actually made a list some time ago about the devices i had my hands on:
dual i2s:
Ball v1
Breadboard
Breadboard Mini (purple)
Guition 1.80 v1
Guition 1.80 v2
Waveshare 1.85 v1
Waveshare 1.85 v2
Waveshare 3.49
Zingzhi Cube 1.54
single i2s:
Lichuang S3 camera
Ball v2
Echoear v1
Echoear v2
Muma horse v1
Muma horse v2
Muma box v1
Muma box v2
Waveshare Watch
Guition 4.3"
Guition 7"
a shame it sort of became standard to use single but nice to hear someone is working on a solution 
there is so many devices out there that it’s impossible to name “the best” and what defines the best, audio quality?, mic quality?, price/value and so on.
the ball been a very popular device partly because it was among the first. it was cheap and it worked
if you want to use it for music playback and audio quality is priority it’s easier as there is only few with audio jack out, respeaker lite, guition 1.80, VoicePE, and DIY, my personal favorite for my desk is guitions 1.80 (aka taichi pi), great quality build, big round display and stereo out.