Today I was super happy, the android app was able to use voice assist without certificate, finally!!!
But bummer, not a single word or command gets understood by fast-piper, always sorry do not understand, while when I type them they work. Maybe I have a terrible speech impediment, in my own language or… Who has a clue how to solve this?
You basically need a GPU to enable accurate STT processing with those larger models. CPU processing is too slow.
how to do that ?
I’m also trying to get this to work with ESPHome and an INM441 without success.
I created a super basic ESPHome YAML to test the INM441 similarly to how koying tested his here: Add more configuration for microphones - i2s/pdm/adc by jesserockz · Pull Request #4775 · esphome/esphome · GitHub right now, mine looks like this:
i2s_audio:
- id: i2s_in
i2s_lrclk_pin: GPIO26
i2s_bclk_pin: GPIO25
microphone:
- platform: i2s_audio
adc_type: external
pdm: false
id: mic_i2s
channel: left
bits_per_sample: 16bit
i2s_audio_id: i2s_in
i2s_din_pin: GPIO33
output:
- platform: gpio
pin: GPIO17
id: light_output
switch:
- platform: gpio
pin: GPIO27
name: L-R Switch
light:
- platform: binary
name: "Response Light"
id: response_light
output: light_output
voice_assistant:
microphone: mic_i2s
binary_sensor:
- platform: gpio
pin:
number: GPIO00
inverted: true
mode:
input: true
pullup: true
name: Boot Switch
internal: true
on_press:
- voice_assistant.start:
- light.turn_on: response_light
on_release:
- voice_assistant.stop:
- light.turn_off: response_light
The device looks like this:
I did try koying’s YAML file for just the microphone too with the same pinouts and it gave the same results. Every time, I see the response: Error: stt-no-text-recognized - No text recognized when trying to issue a voice command.
channel: left to channel: right.pdm: true.I’m running out of things to try. 
Not really of much help. But I also ran out of ideas many months ago and have not tried since. All I could conclude was the INM441 mic was not compatible.
I JUST GOT IT WORKING!!
I bought a set of 5 of these INMP441 microphones. I just built up another (by soldering the pins on it) and swapped it out with the one I had been using for testing and VOILA:
Here’s the ESPHome USB-to-serial debug log data:
[17:55:27][D][binary_sensor:036]: 'Boot Switch': Sending state ON
[17:55:27][D][voice_assistant:132]: Requesting start...
[17:55:27][D][light:036]: 'Response Light' Setting:
[17:55:27][D][light:047]: State: ON
[17:55:27][D][voice_assistant:111]: Starting...
[17:55:27][D][voice_assistant:154]: Assist Pipeline running
[17:55:28][D][sensor:094]: 'Testing Uptime Raw': Sending state 2099.45703 s with 0 decimals of accuracy
[17:55:29][D][binary_sensor:036]: 'Boot Switch': Sending state OFF
[17:55:29][D][voice_assistant:144]: Signaling stop...
[17:55:29][D][light:036]: 'Response Light' Setting:
[17:55:29][D][light:047]: State: OFF
[17:55:30][D][voice_assistant:168]: Speech recognised as: " Turn on the office light."
[17:55:30][D][voice_assistant:144]: Signaling stop...
[17:55:30][D][voice_assistant:192]: Response: "Turned on light"
[17:55:30][D][voice_assistant:207]: Response URL: "http://[REDACTED]:8123/api/tts_proxy/c9423eae01959b2af87c0b8d21f861b36e9b0fec_en-us_8f9b84fea8_tts.piper.raw"
[17:55:30][D][voice_assistant:218]: Assist Pipeline ended
Success! I actually found out that you can add something like this to your configuration.yaml file to capture the WAV files from your ESPHome microphone:
assist_pipeline:
debug_recording_dir: /share/assist_pipeline
That worked as well. I now have a recording of me saying “Turn off the office light”. It sounds pretty bad so I need to figure out how to clean it up but at least it was clear enough that HA understood what I wanted!
Make sure you use 32bit mode for the INMP441 – the recording is incredibly clearer with 32bit rather than 16bit mode:
microphone:
- platform: i2s_audio
adc_type: external
pdm: false
id: mic_i2s
channel: right
bits_per_sample: 32bit
i2s_din_pin: GPIO33
This is how mine looks, for reference.
Hi SpikeyGG !
Could you share the yaml file ?
Thanks
Sure… It’s the same as my post from 6 days ago but with 32bit instead of 16bit:
i2s_audio:
- id: i2s_in
i2s_lrclk_pin: GPIO26
i2s_bclk_pin: GPIO25
microphone:
- platform: i2s_audio
adc_type: external
pdm: false
id: mic_i2s
channel: left
bits_per_sample: 32bit
i2s_audio_id: i2s_in
i2s_din_pin: GPIO33
output:
- platform: gpio
pin: GPIO17
id: light_output
switch:
- platform: gpio
pin: GPIO27
name: L-R Switch
light:
- platform: binary
name: "Response Light"
id: response_light
output: light_output
voice_assistant:
microphone: mic_i2s
binary_sensor:
- platform: gpio
pin:
number: GPIO00
inverted: true
mode:
input: true
pullup: true
name: Boot Switch
internal: true
on_press:
- voice_assistant.start:
- light.turn_on: response_light
on_release:
- voice_assistant.stop:
- light.turn_off: response_light
you can use the ‘BOOT’ button to turn on the STT. You can use the software switch called ‘L-R Switch’ to change the microphone orientation - use this as a debug if you don’t get any audio. The idea is you’re trying to line up the channel: left with what the microphone is sending.
Thanks !
Waiting my INMP441 for testing 
Have a good day
Also wating for some INMP441’s, stupid question will this all work with D1-mini’s? thanks take care
I don’t see why it wouldn’t work. However, I don’t think the D1 mini has any DACs built in so you probably couldn’t do a stereo speaker output easily like with an ESP32, if you wanted the device to use TTS and respond.
I’m trying to debug my voice assist and getting nowhere, in fact I’m in a worse state now than when I started! Anyone got a log output after the latest updates to HA and ESPHome?
Here’s mine, which I think is saying my voice assist pipeline is not receiving anything at all! My assist can control my lights as expected, I just can’t get any STT to it.
[14:40:03][D][binary_sensor:036]: 'Talk Button': Sending state ON
[14:40:03][D][voice_assistant:366]: State changed from IDLE to START_PIPELINE
[14:40:03][D][voice_assistant:372]: Desired state set to START_MICROPHONE
[14:40:03][D][light:036]: 'RGB LED' Setting:
[14:40:03][D][light:047]: State: ON
[14:40:22][D][binary_sensor:036]: 'Talk Button': Sending state OFF
[14:40:22][D][voice_assistant:366]: State changed from START_PIPELINE to STOP_MICROPHONE
[14:40:22][D][voice_assistant:372]: Desired state set to IDLE
[14:40:22][D][light:036]: 'RGB LED' Setting:
[14:40:22][D][light:047]: State: OFF
I have an ESP32 Dev board, INMP441, MAX98357A (and small speaker), WS2812 and switch. Whisper is configured and enabled, as is Piper and OpenWakeWord. Voice Assistant is set for my HA and all fields have (what I think) sensible entries.
If I use some Arduino code to test my microphone and speaker, they seem to work. I’ve added the config to save the recordings as suggested by @SpikeyGG above and I get folders being created but they are empty. I’ve tried changing Left/Right, 16/32bit but am getting nowhere and wishing I got a M5Stack ATOM Echo when they were available last week!
I also get no response from my esp32 dev board with INMP441. I tried various settings. My M5stack echo though is working perfectly now after todays Esphome update, and was working before the update with the no response issue a lot had.
@Arh Curious isn’t it? As I said it appears as though both my microphone and speaker work, but voice assistant doesn’t like them. I wonder if it’s the ESP32 Dev board??? I don’t have a spare board of any type at the moment to test with unfortunately. Been holding off getting whilst I make my mind what to go for… But that’s for another post!
I know I said “for another post”, BUT what boards are people using in their Voice Assistant builds? That are working!
On my board there’s a ESP32-WROOM32 module and I’m using
i2s_lrclock: GPIO25
i2s_bclock: GPIO33
i2s_din: GPIO32
i2s_dout: GPIO14
onboard_led: GPIO02
onboard_button: GPIO00
rgb_led: GPIO13
push_to_talk: GPIO23
Finally got it working, well mostly! The issue with the board I have is that the i2s clock NEEDS to be on certain pins, when I used one of these pins, I have voice! In my case I used GPIO3, all other pins were unchanged.
Not yet perfect, but it works!
Documentation:
https://www.espressif.com/sites/default/files/documentation/esp32-wroom-32_datasheet_en.pdf
https://www.espressif.com/sites/default/files/documentation/esp32_datasheet_en.pdf
Using the following from @SpikeyGG I’m able to hear what was sent from STT
assist_pipeline:
debug_recording_dir: /share/assist_pipeline
Do you mean the L-R switch GPIO should be connected to the L/R of the microphone so as to allow switching left or right?
That’s what I did, simply for debug.
Where did you get the updated version from? I saw the bugfix on github, but is this now available via the webinstaller, or do I have to have a local beta version of esphome installed for the compiling/flashing? Thanks in advance 