Hi community!
Is it possible to record sound from I2S microphone which connected to ESPHome, and save this record i.e to media folder?
1 Like
hello,
i thought about the same thing today 
i have a decibel-meter already with esp32 and i2s mic.
i think its possible because Assist already record a query and convert it to text.
I’m wondering how to do anything with i²s. I bought a set of i2s microphones and I’ve been trying to get them to work with ESPHome but I haven’t had any luck.
I created this thread a while back and no one has responded to it. 
Hi there!
you need a INMP441 : https://fr.aliexpress.com/item/1005004789925958.html?spm=a2g0o.order_list.order_list_main.49.5d235e5b0WjKGX&gatewayAdapt=glo2fra
and any ESP32 cause doesnt work with an ESP8266.
its working like a charm with this config :
sphome:
name: decibel-meter-advanced
friendly_name: decibel-meter-advanced
esp32:
board: wemos_d1_mini32
framework:
type: arduino
external_components:
- source: github://stas-sl/esphome-sound-level-meter
# Enable logging
logger:
level: DEBUG
# Enable Home Assistant API
api:
encryption:
key: "exxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx="
ota:
password: "dxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxf"
wifi:
ssid: !secret wifi_ssid
password: !secret wifi_pwd
manual_ip:
static_ip: 192.168.1.xxx
gateway: 192.168.1.xxx
subnet: 255.255.255.0
dns1: 192.168.1.xxx
fast_connect: on
power_save_mode: none
i2s:
bck_pin: GPIO5
ws_pin: GPIO25
din_pin: GPIO26
sample_rate: 48000 # default: 48000
bits_per_sample: 32 # default: 32
dma_buf_count: 8 # default: 8
dma_buf_len: 256 # default: 256
use_apll: true # default: false
# right shift samples.
# for example if mic has 24 bit resolution, and
# i2s configured as 32 bits, then audio data will be aligned left (MSB)
# and LSB will be padded with zeros, so you might want to shift them right by 8 bits
bits_shift: 8 # default: 0
sound_level_meter:
id: sound_level_meter1
# update_interval specifies over which interval to aggregate audio data
# you can specify default update_interval on top level, but you can also override
# it further by specifying it on sensor level
update_interval: 60s # default: 60s
# you can disable (turn off) component by default (on boot)
# and turn it on later when needed via sound_level_meter.turn_on/toggle actions
is_on: true # default: true
# buffer_size is in samples (not bytes), so for float data type
# number of bytes will be buffer_size * 4
buffer_size: 1024 # default: 1024
# ignore audio data at startup for this long
warmup_interval: 500ms # default: 500ms
# audio processing runs in a separate task, you can change its settings below
task_stack_size: 4096 # default: 4096
task_priority: 2 # default: 2
task_core: 1 # default: 1
# see your mic datasheet to find sensitivity and reference SPL.
# those are used to convert dB FS to db SPL
mic_sensitivity: -26dB # default: empty
mic_sensitivity_ref: 94dB # default: empty
# additional offset if needed
offset: 0dB # default: empty
# for flexibility sensors are organized hierarchically into groups. each group
# could have any number of filters, sensors and nested groups.
# for examples if there is a top level group A with filter A and nested group B
# with filter B, then for sensors inside group B filters A and then B will be
# applied:
# groups:
# # group A
# - filters:
# - filter A
# groups:
# # group B
# - filters:
# - filter B
# sensors:
# - sensor X
groups:
# group 1 (mic eq)
- filters:
# for now only SOS filter type is supported, see math/filter-design.ipynb
# to learn how to create or convert other filter types to SOS
- type: sos
coeffs:
# INMP441:
# b0 b1 b2 a1 a2
- [ 1.0019784 , -1.9908513 , 0.9889158 , -1.9951786 , 0.99518436]
# nested groups
groups:
# group 1.1 (no weighting)
- sensors:
# 'eq' type sensor calculates Leq (average) sound level over specified period
- type: eq
name: LZeq_1s
id: LZeq_1s
# you can override updated_interval specified on top level
# individually per each sensor
update_interval: 1s
# you can have as many sensors of same type, but with different
# other parameters (e.g. update_interval) as needed
- type: eq
name: LZeq_1min
id: LZeq_1min
unit_of_measurement: dBZ
# 'max' sensor type calculates Lmax with specified window_size.
# for example, if update_interval is 60s and window_size is 1s
# then it will calculate 60 Leq values for each second of audio data
# and the result will be max of them
- type: max
name: LZmax_1s_1min
id: LZmax_1s_1min
window_size: 1s
unit_of_measurement: dBZ
# same as 'max', but 'min'
- type: min
name: LZmin_1s_1min
id: LZmin_1s_1min
window_size: 1s
unit_of_measurement: dBZ
# it finds max single sample over whole update_interval
- type: peak
name: LZpeak_1min
id: LZpeak_1min
unit_of_measurement: dBZ
# group 1.2 (A-weighting)
- filters:
# for now only SOS filter type is supported, see math/filter-design.ipynb
# to learn how to create or convert other filter types to SOS
- type: sos
coeffs:
# A-weighting:
# b0 b1 b2 a1 a2
- [ 0.16999495 , 0.741029 , 0.52548885 , -0.11321865 , -0.056549273]
- [ 1. , -2.00027 , 1.0002706 , -0.03433284 , -0.79215795 ]
- [ 1. , -0.709303 , -0.29071867 , -1.9822421 , 0.9822986 ]
sensors:
- type: eq
name: LAeq_1min
id: LAeq_1min
unit_of_measurement: dBA
- type: max
name: LAmax_1s_1min
id: LAmax_1s_1min
window_size: 1s
unit_of_measurement: dBA
- type: min
name: LAmin_1s_1min
id: LAmin_1s_1min
window_size: 1s
unit_of_measurement: dBA
- type: peak
name: LApeak_1min
id: LApeak_1min
unit_of_measurement: dBA
# group 1.3 (C-weighting)
- filters:
# for now only SOS filter type is supported, see math/filter-design.ipynb
# to learn how to create or convert other filter types to SOS
- type: sos
coeffs:
# C-weighting:
# b0 b1 b2 a1 a2
- [-0.49651518 , -0.12296628 , -0.0076134163, -0.37165618 , 0.03453208 ]
- [ 1. , 1.3294908 , 0.44188643 , 1.2312505 , 0.37899444 ]
- [ 1. , -2. , 1. , -1.9946145 , 0.9946217 ]
sensors:
- type: eq
name: LCeq_1min
id: LCeq_1min
unit_of_measurement: dBC
- type: max
name: LCmax_1s_1min
id: LCmax_1s_1min
window_size: 1s
unit_of_measurement: dBC
- type: min
name: LCmin_1s_1min
id: LCmin_1s_1min
window_size: 1s
unit_of_measurement: dBC
- type: peak
name: LCpeak_1min
id: LCpeak_1min
unit_of_measurement: dBC
# automation
# available actions:
# - sound_level_meter.turn_on
# - sound_level_meter.turn_off
# - sound_level_meter.toggle
switch:
- platform: template
name: "Sound Level Meter Switch"
lambda: |-
return id(sound_level_meter1).is_on();
turn_on_action:
then: sound_level_meter.turn_on
turn_off_action:
then: sound_level_meter.turn_off
restore_mode: RESTORE_DEFAULT_ON
- platform: restart
name: "Restart Decibel Metre Advance"
status_led:
pin:
number: GPIO2 #ESP32 OnBroad LED
inverted: true
binary_sensor:
- platform: status
name: "Status Wifi Decibel Metre Advance"
sensor:
- platform: internal_temperature
name: "Internal Temperature Decibel Metre"
ENJOY 
@Joe3
Nah, it works. I have both types of i²s sensors actually. Eventually, I got some responses and we figured it out.
Thanks for posting.
It also works with a simple config (YAML) which I found here https://github.com/stas-sl/esphome-sound-level-meter
esphome:
name: esp32-02
platform: esp32
board: esp32dev
external_components:
- source: github://stas-sl/esphome-sound-level-meter
# Enable logging
logger:
level: DEBUG
# Enable Home Assistant API
api:
encryption:
key: " xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
ota:
password: "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
wifi:
ssid: !secret wifi_ssid
password: !secret wifi_password
fast_connect: true
manual_ip:
static_ip: 192.168.178.xx
gateway: 192.168.178.x
subnet: 255.255.255.0
# Enable fallback hotspot (captive portal) in case wifi connection fails
ap:
ssid: "Esp32-02 Fallback Hotspot"
password: "eVv4rVmMxxfi"
captive_portal:
# ---------------------------
# INMP441 I2S MEMS microphone
# VDD 3V3
# GND GND
# SD D26
# L/R GND
# WS D25
# SCK D5
status_led:
pin: 2
i2s:
bck_pin: GPIO5 # SCK
ws_pin: GPIO25 # WS
din_pin: GPIO26 # SD
sample_rate: 48000 # default: 48000
bits_per_sample: 32 # default: 32
# right shift samples.
# for example if mic has 24 bit resolution, and i2s configured as 32 bits,
# then audio data will be aligned left (MSB) and LSB will be padded with
# zeros, so you might want to shift them right by 8 bits
bits_shift: 8 # default: 0
sound_level_meter:
id: sound_level_meter1
# update_interval specifies over which interval to aggregate audio data
# you can specify default update_interval on top level, but you can also
# override it further by specifying it on sensor level
update_interval: 1s # default: 60s
# buffer_size is in samples (not bytes), so for float data type
# number of bytes will be buffer_size * 4
buffer_size: 1024 # default: 1024
# see your mic datasheet to find sensitivity and reference SPL.
# those are used to convert dB FS to db SPL
mic_sensitivity: -26dB # default: empty
mic_sensitivity_ref: 94dB # default: empty
# for flexibility sensors are organized hierarchically into groups.
# each group can have any number of filters, sensors and nested groups.
# for examples if there is a top level group A with filter A and nested
# group B with filter B, then for sensors inside group B filters A
# and then B will be applied:
# groups:
# # group A
# - filters:
# - filter A
# groups:
# # group B
# - filters:
# - filter B
# sensors:
# - sensor X
groups:
- sensors:
- type: eq
name: Leq_1s
# automation
# available actions:
# - sound_level_meter.turn_on
# - sound_level_meter.turn_off
# - sound_level_meter.toggle
switch:
- platform: template
name: "Decibel Meter ON / OFF"
lambda: |-
return id(sound_level_meter1).is_on();
turn_on_action:
then: sound_level_meter.turn_on
turn_off_action:
then: sound_level_meter.turn_off
restore_mode: RESTORE_DEFAULT_ON
- platform: restart
name: "Decibel Meter Reset"
sensor:
- platform: wifi_signal
name: "WiFi ESP32 02"
update_interval: 60s
- platform: internal_temperature
name: "Decibel Meter Temperature"
binary_sensor:
- platform: status
name: "Decibel Meter WiFi"
I added the ON-OFF switch, internal temperature, RESET-switch and WiFi-status from the advanced script to the simple script which worked without any failures.
Posting your config would be helpful 
OK, so esphome-sound-level-meter is working like a charm right now, but is there a way to record / stream the captured audio to Home Assistant to be played in any media player entity?
Thanks in advance
I’m trying to implement this sound meter code for the m5 atom echo which has an I2S microphone as I’m trying to troubleshoot if the embedded microphone is not working at all or what (towards usage as voice assistant).
I tried to map the GPIO from the atom (ref1, ref2) to those in the sound meter template but I’m unsure of what to map to each variable needed.
While it compiles and runs the code, the output is always -inf in the logs, so not sure whether I did wrongly apply the GPIOs, or something else.
[13:23:51][D][sound_level_meter:120][sound_level_met]: Processing time per 1s of audio data (48000 samples): 21 ms
[13:23:52][D][sensor:094]: 'Leq_1s': Sending state -inf dB with 2 decimals of accuracy
So can’t tell if microphone is, as presumed, bad in this unit, or the code is wrong somewhere, or there’s some incompatibility between the sound meter library and the atom lite base.
- Any pointers on what to look for?
- Anyone with success on this platform?
Thanks!
Code
substitutions:
name: atom-sound-meter
friendly_name: Sound Meter
esphome:
name: ${name}
friendly_name: ${friendly_name}
name_add_mac_suffix: true
project:
name: m5stack.atom-echo-sound-meter
version: "1.0.1"
min_version: 2023.11.1
esp32:
board: m5stack-atom
framework:
type: esp-idf
wifi:
ssid: !mySSID
password: !myPassword
use_address: !myIP
on_connect:
- delay: 5s # Gives time for improv results to be transmitted
- ble.disable:
on_disconnect:
- ble.enable:
api:
logger:
ota:
safe_mode: true
improv_serial:
esp32_improv:
authorizer: none
button:
- platform: factory_reset
id: factory_reset_btn
name: Factory reset
external_components:
- source: github://stas-sl/esphome-sound-level-meter
- source: github://pr#5230
components:
- esp_adf
refresh: 0s
esp_adf:
status_led:
pin: 27
i2s:
bck_pin: GPIO19 # SCK
ws_pin: GPIO33 # WS
din_pin: GPIO23 # SD
sample_rate: 48000 # default: 48000
bits_per_sample: 32 # default: 32
# right shift samples.
# for example if mic has 24 bit resolution, and i2s configured as 32 bits,
# then audio data will be aligned left (MSB) and LSB will be padded with
# zeros, so you might want to shift them right by 8 bits
bits_shift: 8 # default: 0
sound_level_meter:
id: sound_level_meter1
# update_interval specifies over which interval to aggregate audio data
# you can specify default update_interval on top level, but you can also
# override it further by specifying it on sensor level
update_interval: 1s # default: 60s
# buffer_size is in samples (not bytes), so for float data type
# number of bytes will be buffer_size * 4
buffer_size: 1024 # default: 1024
# see your mic datasheet to find sensitivity and reference SPL.
# those are used to convert dB FS to db SPL
mic_sensitivity: -22dB # default: empty
mic_sensitivity_ref: 94dB # default: empty
groups:
- sensors:
- type: eq
name: Leq_1s
switch:
- platform: template
name: "Decibel Meter ON / OFF"
lambda: |-
return id(sound_level_meter1).is_on();
turn_on_action:
then: sound_level_meter.turn_on
turn_off_action:
then: sound_level_meter.turn_off
restore_mode: RESTORE_DEFAULT_ON
- platform: restart
name: "Decibel Meter Reset"
sensor:
- platform: wifi_signal
name: "WiFi ESP32 02"
update_interval: 60s
- platform: internal_temperature
name: "Decibel Meter Temperature"
binary_sensor:
- platform: status
name: "Decibel Meter WiFi Status"
What about waterfall visualization?
Real time fft is cool but sometimes events happens when we’re not looking at screen so whould be great to have a feature like this.
Maybe showing latest x secs or hours, and streaming everything to hassio to store it