M5Stack Atom Echo stops listening after one voice command

[Taomyn]

The problem

Since the last two ESPHome updates, 2025.6.0 and 2025,5,2, my Atom Echo only responds to one voice command after which it never responds again until the device is rebooted. I know it's not hung because I can press the button on the device and the log responds telling me that it's already listening

Which device are you using?

m5stack-atom-echo

Firmware Version

Firmware: 2025.6.0 (Jun 18 2025, 19:36:18)

Logs

INFO ESPHome 2025.6.0
INFO Reading configuration /config/esphome/m5ae-31b338.yaml...
INFO Generating C++ source...
INFO Updating https://github.com/espressif/esp-protocols.git@mdns-v1.8.2
INFO Updating https://github.com/espressif/esp-tflite-micro@v1.3.3.1
INFO Compiling app...
Processing m5ae-31b338 (board: m5stack-atom; framework: espidf; platform: https://github.com/pioarduino/platform-espressif32/releases/download/53.03.13/platform-espressif32.zip)
--------------------------------------------------------------------------------
HARDWARE: ESP32 240MHz, 320KB RAM, 4MB Flash
 - framework-espidf @ 3.50302.0 (5.3.2) 
 - tool-cmake @ 3.30.2 
 - tool-esptoolpy @ 4.8.6 
 - tool-mklittlefs @ 3.2.0 
 - tool-ninja @ 1.7.1 
 - tool-riscv32-esp-elf-gdb @ 14.2.0+20240403 
 - tool-xtensa-esp-elf-gdb @ 14.2.0+20240403 
 - toolchain-esp32ulp @ 2.35.0-20220830 
 - toolchain-xtensa-esp-elf @ 13.2.0+20240530
Reading CMake configuration...
Dependency Graph
|-- noise-c @ 0.1.6
|-- ArduinoJson @ 6.18.5
|-- ESPMicroSpeechFeatures @ 1.1.0
|-- esp-audio-libs @ 1.1.4
Compiling .pioenvs/m5ae-31b338/src/main.cpp.o
Linking .pioenvs/m5ae-31b338/firmware.elf
RAM:   [=         ]  11.5% (used 37640 bytes from 327680 bytes)
Flash: [========= ]  88.5% (used 1624120 bytes from 1835008 bytes)
Building .pioenvs/m5ae-31b338/firmware.bin
Creating esp32 image...
Successfully created esp32 image.
esp32_create_combined_bin([".pioenvs/m5ae-31b338/firmware.bin"], [".pioenvs/m5ae-31b338/firmware.elf"])
SHA digest in image updated
Wrote 0x19c9a0 bytes to file /data/build/m5ae-31b338/.pioenvs/m5ae-31b338/firmware.factory.bin, ready to flash to offset 0x0
esp32_copy_ota_bin([".pioenvs/m5ae-31b338/firmware.bin"], [".pioenvs/m5ae-31b338/firmware.elf"])
========================= [SUCCESS] Took 16.98 seconds =========================
INFO Successfully compiled program.
INFO Connecting to 192.168.1.39 port 3232...
INFO Connected to 192.168.1.39
INFO Uploading /data/build/m5ae-31b338/.pioenvs/m5ae-31b338/firmware.bin (1624480 bytes)
Uploading: [============================================================] 100% Done...

INFO Upload took 6.35 seconds, waiting for result...
INFO OTA successful
INFO Successfully uploaded program.
INFO Starting log output from 192.168.1.39 using esphome API
INFO Successfully resolved m5ae-31b338 @ 192.168.1.39 in 0.000s
INFO Successfully connected to m5ae-31b338 @ 192.168.1.39 in 5.142s
INFO Successful handshake with m5ae-31b338 @ 192.168.1.39 in 0.112s
[11:15:25][I][app:137]: ESPHome version 2025.6.0 compiled on Jun 19 2025, 11:14:54
[11:15:25][C][wifi:613]: WiFi:
[11:15:25][C][wifi:434]:   Local MAC: [redacted]
[11:15:25][C][wifi:439]:   SSID: [redacted]
[11:15:25][C][wifi:442]:   IP Address: 192.168.1.39
[11:15:25][C][wifi:446]:   BSSID: [redacted]
[11:15:25][C][wifi:446]:   Hostname: 'm5ae-31b338'
[11:15:25][C][wifi:446]:   Signal strength: -44 dB ▂▄▆█
[11:15:25][C][wifi:455]:   Channel: 1
[11:15:25][C][wifi:455]:   Subnet: 255.255.255.0
[11:15:25][C][wifi:455]:   Gateway: 192.168.1.1
[11:15:25][C][wifi:455]:   DNS1: 192.168.1.11
[11:15:25][C][wifi:455]:   DNS2: 192.168.1.71
[11:15:25][D][api.connection:1570]: Home Assistant 2025.6.1 (192.168.1.18) connected
[11:15:25][C][logger:211]: Logger:
[11:15:25][C][logger:211]:   Max Level: DEBUG
[11:15:25][C][logger:211]:   Initial Level: DEBUG
[11:15:25][C][logger:217]:   Log Baud Rate: 115200
[11:15:25][C][logger:217]:   Hardware UART: UART0
[11:15:25][C][logger:224]:   Task Log Buffer Size: 768
[11:15:25][C][esp32_rmt_led_strip:250]: ESP32 RMT LED Strip:
[11:15:25][C][esp32_rmt_led_strip:250]:   Pin: 27
[11:15:25][C][esp32_rmt_led_strip:255]:   RMT Symbols: 192
[11:15:25][C][esp32_rmt_led_strip:283]:   RGB Order: GRB
[11:15:25][C][esp32_rmt_led_strip:283]:   Max refresh rate: 0
[11:15:25][C][esp32_rmt_led_strip:283]:   Number of LEDs: 1
[11:15:25][C][template.select:065]: Template Select 'Wake word engine location'
[11:15:25][C][template.select:066]:   Update Interval: 60.0s
[11:15:25][C][template.select:069]:   Optimistic: YES
[11:15:25][C][template.select:069]:   Initial Option: On device
[11:15:25][C][template.select:069]:   Restore Value: YES
[11:15:25][C][gpio.binary_sensor:015]: GPIO Binary Sensor 'Button'
[11:15:25][C][gpio.binary_sensor:016]:   Pin: GPIO39
[11:15:25][C][light:092]: Light 'M5Stack Atom Echo 31b338'
[11:15:25][C][light:094]:   Default Transition Length: 0.0s
[11:15:25][C][light:094]:   Gamma Correct: 2.80
[11:15:25][C][template.switch:079]: Template Switch 'Use listen light'
[11:15:25][C][template.switch:079]:   Restore Mode: restore defaults to ON
[11:15:25][C][template.switch:057]:   Optimistic: YES
[11:15:25][C][template.switch:079]: Template Switch 'timer_ringing'
[11:15:25][C][template.switch:079]:   Restore Mode: always OFF
[11:15:25][C][template.switch:057]:   Optimistic: YES
[11:15:25][C][factory_reset.button:011]: Factory Reset Button 'Factory reset'
[11:15:25][C][factory_reset.button:011]:   Icon: 'mdi:restart-alert'
[11:15:25][C][i2s_audio.microphone:083]: Microphone:
[11:15:25][C][i2s_audio.microphone:083]:   Pin: 23
[11:15:25][C][i2s_audio.microphone:083]:   PDM: YES
[11:15:25][C][i2s_audio.microphone:083]:   DC offset correction: YES
[11:15:25][C][psram:018]: PSRAM:
[11:15:25][C][psram:022]:   Available: NO
[11:15:26][C][i2s_audio.speaker:114]: Speaker:
[11:15:26][C][i2s_audio.speaker:114]:   Pin: 22
[11:15:26][C][i2s_audio.speaker:114]:   Buffer duration: 60
[11:15:26][C][i2s_audio.speaker:120]:   Timeout: 500 ms
[11:15:26][C][i2s_audio.speaker:128]:   Communication format: std
[11:15:26][C][captive_portal:089]: Captive Portal:
[11:15:26][C][esphome.ota:073]: Over-The-Air updates:
[11:15:26][C][esphome.ota:073]:   Address: 192.168.1.39:3232
[11:15:26][C][esphome.ota:073]:   Version: 2
[11:15:26][C][safe_mode:018]: Safe Mode:
[11:15:26][C][safe_mode:019]:   Boot considered successful after 60 seconds
[11:15:26][C][safe_mode:019]:   Invoke after 10 boot attempts
[11:15:26][C][safe_mode:019]:   Remain for 300 seconds
[11:15:26][D][esp32.preferences:142]: Writing 2 items: 2 cached, 0 written, 0 failed
[11:15:26][C][api:182]: API Server:
[11:15:26][C][api:182]:   Address: 192.168.1.39:6053
[11:15:26][C][api:187]:   Using noise encryption: YES
[11:15:26][C][http_request:013]: HTTP Request:
[11:15:26][C][http_request:013]:   Timeout: 4500ms
[11:15:26][C][http_request:013]:   User-Agent: ESPHome/2025.6.0 (https://esphome.io)
[11:15:26][C][http_request:013]:   Follow redirects: YES
[11:15:26][C][http_request:013]:   Redirect limit: 3
[11:15:26][C][http_request.idf:029]:   Buffer Size RX: 512
[11:15:26][C][http_request.idf:029]:   Buffer Size TX: 512
[11:15:26][C][mdns:122]: mDNS:
[11:15:26][C][mdns:122]:   Hostname: m5ae-31b338
[11:15:26][C][micro_wake_word:064]: microWakeWord:
[11:15:26][C][micro_wake_word:065]:   models:
[11:15:26][C][micro_wake_word:014]:     - Wake Word: Okay Nabu
[11:15:26][C][micro_wake_word:014]:       Probability cutoff: 0.97
[11:15:26][C][micro_wake_word:014]:       Sliding window size: 5
[11:15:26][C][micro_wake_word:014]:     - Wake Word: Hey Mycroft
[11:15:26][C][micro_wake_word:014]:       Probability cutoff: 0.95
[11:15:26][C][micro_wake_word:014]:       Sliding window size: 5
[11:15:26][C][micro_wake_word:014]:     - Wake Word: Hey Jarvis
[11:15:26][C][micro_wake_word:014]:       Probability cutoff: 0.97
[11:15:26][C][micro_wake_word:014]:       Sliding window size: 5
[11:15:26][C][micro_wake_word:022]:     - VAD Model
[11:15:26][C][micro_wake_word:022]:       Probability cutoff: 0.50
[11:15:26][C][micro_wake_word:022]:       Sliding window size: 5
[11:15:28][D][micro_wake_word:360]: Starting wake word detection
[11:15:28][D][micro_wake_word:378]: State changed from STOPPED to STARTING
[11:15:28][D][ring_buffer:034][mww]: Created ring buffer with size 3840
[11:15:28][D][micro_wake_word:261]: Inference task has started, attempting to allocate memory for buffers
[11:15:28][D][micro_wake_word:266]: Inference task is running
[11:15:28][D][micro_wake_word:378]: State changed from STARTING to DETECTING_WAKE_WORD
[11:16:20][I][safe_mode:042]: Boot seems successful; resetting boot loop counter
[11:16:20][D][esp32.preferences:142]: Writing 1 items: 0 cached, 1 written, 0 failed

Additional information

substitutions:
  name: "m5ae-31b338"
  friendly_name: M5Stack Atom Echo 31b338

packages:
  base: github://esphome/wake-word-voice-assistants/m5stack-atom-echo/m5stack-atom-echo.yaml@main

esphome:
  name: ${name}
  name_add_mac_suffix: false
  friendly_name: ${friendly_name}

api:
  encryption:
    key: !secret m5ae_key

wifi:
  ssid: !secret wifi_ssid
  password: !secret wifi_password
  manual_ip:
    static_ip: 192.168.1.39
    subnet: 255.255.255.0
    gateway: 192.168.1.1
    dns1: 192.168.1.11
    dns2: 192.168.1.71
  domain: !secret dns_suffix
  on_connect:
    - delay: 5s  # Gives time for improv results to be transmitted

http_request:
  verify_ssl: false

micro_wake_word:
  on_wake_word_detected:
    then:
      - media_player.speaker.play_on_device_media_file:
          media_file: timer_finished_wave_file
          announcement: true
      - delay: 10ms
      - wait_until:
          - media_player.is_idle:
          - speaker.is_stopped:
      - voice_assistant.start:
          wake_word: !lambda return wake_word;

media_player:
  - id: !extend echo_media_player
    on_idle:
      then:
        - delay: 1ms

I have tried removing some of the "additions" in my yaml e.g. the whole micro_wake_word section, the on_connect parameter, media_player section, but nothing helps and sometimes makes it worse and it doesn't respond at all.

[Subcode]

I'm having the same issue, tried using the new example but doesn't work.

[spudje]

I have similar issues:

• It doesn't do continued conversation anymore, i.e. doesn't listen for a response after posing a question
• I have to wait 10 to 20 seconds after an answer before it starts detecting the wake word again.

[benson789]

Same issue here on 3 different atom echos.

***Edit - my pipeline was broken. My bad. Same symptoms as above though. ***

[danuw]

I get this error when I use the yaml from line 29 (after cap) at https://github.com/esphome/wake-word-voice-assistants/blob/main/m5stack-atom-echo/m5stack-atom-echo.yaml to replace mine

VAD threshold is no longer supported, as it requires the deprecated esp_adf external component. Use an i2s_audio microphone/speaker instead. Additionally, you may need to configure the audio_adc and audio_dac components depending on your hardware.
enum: Use an external DAC, for example the NS4168, or UDA1334A.

internal: Use the internal DAC

Anyone else seeing this (I just updated after a couple of months)

[alexburkert]

I am experiencing the same issue.
For the sake of asking, has anyone arrived at a workaround?
Thank you,
Burkert

[Delfairen]

I get this error when I use the yaml from line 29 (after cap) at https://github.com/esphome/wake-word-voice-assistants/blob/main/m5stack-atom-echo/m5stack-atom-echo.yaml to replace mine

VAD threshold is no longer supported, as it requires the deprecated esp_adf external component. Use an i2s_audio microphone/speaker instead. Additionally, you may need to configure the audio_adc and audio_dac components depending on your hardware.
enum: Use an external DAC, for example the NS4168, or UDA1334A.

internal: Use the internal DAC

Anyone else seeing this (I just updated after a couple of months)

I am also getting this same error

[bcutter]

It’s a memory issue isn’t it? Saw something related in the logs which „might impact micro word detection“ so…?

[alexburkert]

I tried the solution at this thread, worked perfectly.

[bcutter]

Well that sounds like a workaround worth looking into (hot trace):

Whatever the setting triggers on the device is the action needed.

(I only see this in the ESP logs when configuring the setting, I don't think we can make much out of that:

[12:42:22][D][select:062]: 'Wake word engine location' - Setting
[12:42:22][D][select:115]: 'Wake word engine location' - Set selected option to: On device
[12:42:22][D][select:015]: 'Wake word engine location': Sending state On device (index 1)
[12:42:23][W][micro_wake_word:353]: Wake word detection is already running
[12:42:41][D][esp32.preferences:142]: Writing 1 items: 1 cached, 0 written, 0 failed

).

How can we natively implement this "out of the box"?

[github-actions]

As there has been no activity on this issue for 30 days, I am marking it as stale. If you think this is a mistake, please comment below and I will remove the stale label.

[bcutter]

Still an issue with only a workaround.