With the default settings of the M5Atom Echo and using the built-in wake.word , I get a “metallic” sound that is unrecognizable. and with long texts, such as more than 10 words, the sound starts to break down, stutter and lose meaning in all sorts of ways.
Are there any parameters you have been able to use to improve the sound on esp32 devices on the i2s bus?
i can try this, and didnt have any result:
speaker:
- platform: i2s_audio
id: echo_speaker
i2s_dout_pin: 22
channel: mono
sample_rate: 16000
bits_per_sample: 8bit
i2s_mode: primary
use_apll: false
bits_per_channel: default
buffer_duration: 1200ms
timeout: 400ms
#i2s_comm_fmt: pcm
dac_type: external
I really like the new option to use the LLM/OpenAI as a second voice command handler, if the intent is not recognized, HomeAssistant now sends the text to my local LLM, and then the response becomes longer than usual and Atom Echo can’t handle it.
Translated with DeepL.com (free version)