I’m having the same issue since the update. It was working fine previously.
2 Likes
I also don’t get spoken responses anymore.
2 Likes
Adding my voice to people who aren’t getting spoken responses from the S3 Box3 after the 2024.5.0 update
3 Likes
Oh happy day, well kinda (happy in that misery loves company and it wasn’t something I did). Finally a thread where someone else sees what I see. Since 2024.5.0, my S3 box 3 can now hear me but it’s silent. 
3 Likes
Can you please tell me exactly where should I insert this yaml code? In a “voice_assistant” section of the esp32-s3-box-3 yaml conf file?
I also did not get any spoken response with 2024.5.0.
reverting esphome CLI to 2024.04.2 worked.
1 Like
Assuming spoken response is fixed, does not having a media player also mean no possibility to sent TTS messages, or is that already possible? I’m not interested in music, but I am in announcements.
1 Like
I just did the second update today and still am not getting a spoken response. Is there something I need to add to the configuration now?
1 Like
Short term fix - downgrading to an older version of esp-idf:
esp32:
board: esp32s3box
flash_size: 16MB
framework:
type: esp-idf
version: 4.4.6
3 Likes
I’ve been looking on how to downgrade. I’ve only installed esphome/voice assistant via the web browser before. Do I need to run esphome on my linux laptop and compile a firmware and put it onto the esp32-s3-box-3? Just trying to find some clear steps of how I can put an older version on it.
any help is appreciated.
I also needed this, but first my microphone didn’t work. I’m unsure what change fixed that, maybe the revert to ESPHome 2024.4.x, but I have a working version now with on-device hot word working, mic, Wi-Fi, everything working GitHub - D3SOX/firmware: Holds firmware configuration files for projects that the ESPHome team provides.
Flashed via web esp home on the first click on Configure Wi-Fi it might give an error but try again it always worked the second time for me.
2 Likes
Yes thats correct.
good evening I saw your firmware it is really great and I have no bugs everything works perfect but it is impossible for me to add in the compilation
github://jaymunro/esphome_firmware/wake-word-voice-assistant/esp32-s3-box-3.yaml@continued-conversation
While github://jaymunro/esphome_firmware/esp32-box-docks/box3-sensor-for-voice-assistant.yaml@sensor_dock.
Installs in esphome compilation works
Here is my config for my brand new ESP32-S3-BOX-3 which “works” except the wake word detection is awful. It works maybe 1 in 10 times and that’s when we’re talking directly into the mic (10cm away) quite loudly. No chance from across the room. Am I doing something wrong? Do I have a unit with a duff mic?
# NOTE: Microphone is broken when using ESPHome 2024.5.x, so use 2024.4.2 docker image for now!
# Global variables
substitutions:
name: ha-voice-1
friendly_name: HA Voice 1
micro_wake_word_model: hey_jarvis
# Main package
packages:
esphome.voice-assistant: github://esphome/firmware/wake-word-voice-assistant/esp32-s3-box-3.yaml@main
# Remove MAC from name
esphome:
name_add_mac_suffix: true
# WiFi credentials
wifi:
ssid: !secret wifi_ssid
password: !secret wifi_password
domain: .not.lan
# Bug fix for speaker not working
esp32:
framework:
version: 4.4.6
DragonQ
I too have a brand new box with the same exact problem, with the only difference that I have not installed esphome, it’s still with factory firmware. I came here searching for possible clues, thought to myself, perhaps if I install a different firmware I could get some ideas as to what’s wrong with it… Now I think espressif is just selling crap. My experience with customer service was… totally bizarre, namely solid denial, what you are saying can’t possibly be true, because it just can’t and that’s that. Didn’t really expect that from espressif, but here I am with a box’o’crap instead of a working product and looking for other more reliable vendors…
Perhaps rather than simply using the github package link you could try copying the code and editing the following bit (or using further substitutions), playing with the settings to see if you can get the mic working better.
voice_assistant:
id: va
microphone: box_mic
speaker: box_speaker
use_wake_word: true
noise_suppression_level: 2
auto_gain: 31dBFS
volume_multiplier: 2.0
vad_threshold: 3
Try playing with the volume multiplier and, noise suppression and vad threshold. Having said that, mine isn’t great either.
FYI, copying the code doesn’t achieve anything, you can override those values after including the package.
I have messed with the volume_multipler and vad_threshold a bit but they don’t seem to help.
1 Like
Which is why I also said:
1 Like
I do have another unit to try so if that has the same issues then I think I’ll have to sadly park this as “not even ready for early adopters”.
Maybe some updates in a few months will improve things but it does seem odd that videos exist of people using this hardware seemingly with ease.
EDIT: Good news, the second unit I bought seems to work way better than the first one, despite both running the same software (same YAML except device name). 3/3 wake words recognised so far, all from a metre away.
I will try to return the first one as faulty unless the shipping is prohibitive, in which case I’ll keep it as a device to play with.
Mine works very well at a distance of about 5m.
Have you removed the foil from the display and microphones?