#ESPHome 2024.11 on M5stack Atom Echo

1 messages · Page 1 of 1 (latest)

warped ginkgo
#

Hi guys!
Is anyone having audio troubles with the latest ESPHome update? I can't set timers because Assist misunderstood the intent, the TTS sounds very bad, and I have the same timer ring sound as the wake sound, but it doesn't work. I tried adding bits_per_sample, sample_rate as documentation says but I can't make it work...

heavy cedar
#

Cc @forest girder

forest girder
#

It sounds like several different (possibly related) issues here, so we need some more information. Did the same yaml work fine before the 2024.11 update? Is the reason you can't set the timer because the microphone quality is worse, and the STT isn't working well? How did you setup the timer and wake word ringing exactly?

warped ginkgo
#

Sure, the device worked properly and had no issues before the update. In fact, if I install the version from the Ready-Made projects page of ESPHome, it works correctly. However, after installing it and HA prompts me to update the device's version, the issue reappears. I have no problems with the microphone because it works fine. After updating, my issue is with the TTS sound quality—it has a kind of metallic, very low-quality sound. This is my configuration file.

heavy cedar
#

Didn't the default sample rate change in 2024.11? 🤔

warped ginkgo
#

I hadn’t used any setup related to sample rate before the 2024.11

heavy cedar
#

Yes, i mean internally in the i2s_audio comppnent. Which means that not touching a config would change the default

warped ginkgo
#

If I understand correctly what you said, it might be the case since, as I mentioned earlier, when I install the default configuration from ESPHome Ready-Made Projects, it works fine. However, when linking it to HA and the system prompts me to update, the issue reappears. I'll try to replicate what I’m describing and will attempt to share it.

warped ginkgo
#

Default configuration, the speaker sounds very good, but timers don't work. When I set one, it tells me the command was received, but when I ask about the timer's status, it says there are no active timers.

image.png
#

That's the entire issue

forest girder
#

I can reproduce the poor audio quality on an Echo, I'll look into today

forest girder
#

I think I'm driving myelf crazier, and I can't tell if I am hallucinating a difference in audio quality as I play around with various settings. Would you mind adding a couple of things to your yaml and let me know if the audio quality is good/bad/different? Let's start with building a firmware with the ESPHome 2024.11 but with ```external_components:

  • source: github://pr#7449
    components: [i2s_audio]``` added to your yaml.
#

As a separate test, take out that yaml block and replace it with ```external_components:

  • source: github://pr#7459
    components: [voice_assistant]```
warped ginkgo
#

Sure @forest girder , I'll try it when I get home. Thanks!!

warped ginkgo
#

I used this block but it didn't solve it. The strange thing, which I can't explain, is why the default configuration https://esphome.io/projects/index.html also gives me the same problem after updating.

  - source: github://pr#5230
    components:
      - esp_adf
    refresh: 0s
  - source: github://jesserockz/esphome-components
  - source: github://pr#7459
    components: [voice_assistant]```
ESPHome
Ready-Made Projects

Install ready-made projects directly on your device.

craggy goblet
warped ginkgo
forest girder
#

There is probably a bug in the new speaker code, so that's why the default configuration is broken after the update. Did you try adding this external component? ```external_components:

  • source: github://pr#7449
    components: [i2s_audio]``` That should use the old speaker code. I'm trying to track down where the issue is, and so if adding that to your config makes it sound okay, then I know for sure it is in the new speaker code.
warped ginkgo
warped ginkgo
#

@forest girder Bro, it worked using:

external_components:
  - source: github://pr#7449
    components: [i2s_audio]

Everything is OK now. Thanks a lot!!

forest girder
#

Well, that fixes it because its using the old code. I'll keep trying to track down why my new speaker code is doing this so its fixed for everyone. Thanks for reporting!

forest girder
#

This was a super annoying one to figure out! Apparently the Echo speaker wants 32 bits per sample (the datasheet is in Chinese, but it still wasn't clear even with Google's automatic translation - I figured this out by a random test). The reason the old code worked was basically by accident; the sample was duplicated to convert from 1 channel to 2 channels. This effectively made it 32 bits per sample (the least significant 16 bits were non-random noise, as they were really a copy of the original sample itself). The current code still didn't work even when configured at 32 bits per sample because I had it reconfigure the I2S bus to use the incoming audio's bits per sample rather than the originally configured value. I have a pretty simple fix, but I want to add a few other small improvements into the branch before I make a PR. I'll let you know when its ready to test it out though, as it would be great to have someone else verify its working well!

forest girder
#

Alright, I made a PR for the low audio quality issue on the Echo. I would appreciate if you could try it out and let me know how it sounds! You need to adjust your speaker configuration to this: ```speaker:

  • platform: i2s_audio
    id: echo_speaker
    i2s_dout_pin: GPIO22
    dac_type: external
    bits_per_sample: 32bit
    channel: rightand addexternal_components:
  • source: github://pr#7916
    components: [i2s_audio]``` to use my new PR (remove any other external components importing i2s_audio).
craggy goblet
#

Don't spoil other's thread. Create your own, paste your config and explain thoroughly what doesn't work and which hardware you use and if you tested speaker separately to make sure that it works.