#automation tts different than assist tts

1 messages · Page 1 of 1 (latest)

gusty eagle
#

Hi all, wondering if anyone else has run into this with HA PE.
I have kokoro running as my tts for assist. I have an automation that says "someone is at the door" when my sensor detects motion. Oddly, if i use home assistant cloud entity, it runs great. But if I choose my kokoro entity, it doesn't make any noise.

is this some kind of bug where the response time for kokoro is too slow and speaker is off? or something else? Anyone else have this happen to them?

prisma spruce
#

Trace. Most probably it errors out. I had it. Don't use Kokoro because of it...

gusty eagle
#

trace?

#

i like the voices from kokoro and it works for the rest of assist...

prisma spruce
# gusty eagle trace?

On automation screen, top right, there are traces. Check that.
It should have the TTS URL in data. Check if that URL contains correct audio.
Kokoro is great.

#

How do you run Kokoro BTW? Docker?

gusty eagle
#

oh i see what happens... in the trace it is calling the old name of the entity...

#

instead of the current name of the entity...

#

that seems to be a home assistant coding problem...if that is truly the problem.. i renamed it within homeassistant to kokoro, but it's still calling it by the previous name of openai...

gusty eagle
prisma spruce
gusty eagle
#

yeah, weird thing is i changed the entity a while ago

#

just recently going through my automations.

#

in the automation speak entities, it doesn't show openai, it just shows my current kokoro...

#

i wonder if the renaming is superficial for human reading, but he underlying call is for the same name. in which case, the trace isn't throwing any errors

#

i am also not in the same room of it so i'm not sure if its working or not lol. lemme double check first...

#

Result:
params:
domain: tts
service: speak
service_data:
cache: true
media_player_entity_id: media_player.home_assistant_voice_09560c_media_player
message: Hi, someone is at the door
entity_id:
- tts.openai
target:
entity_id:
- tts.openai
running_script: false

#

this is what shows in the trace, still not playing the voice

prisma spruce
#

but it's still openai

gusty eagle
#

i think that's ok, thats the name being served from the wyoming protocol docker:

#

it is the entity ID in the wyoming protocol integration. so i'm guessing it's targeting the entity id versus the human readable name that i changed

#

i'd imagine that's not the problem...

#

which there was a more indepth trace. i.e. showing that the message was sent via wyoming protocol via the entity id, and then seeing it come back...

#

found the error in the raw logs!:
2025-04-25 12:03:47.778 ERROR (MainThread) [homeassistant.components.esphome.manager] Home Assistant Voice 09560c: [E][speaker_media_player:339]: The announcement pipeline's file reader encountered an error.

prisma spruce
#

Ah, yes, of course then. IDK why your entity has ID "openai" and name "Kokoro", it's super confusing, but the automation is right.
Looks like Kokoro doesn't generate files properly. It happened to me too. You have to check the voice you're trying to use.

gusty eagle
#

did you get it to work?

#

kokoro works fine for voice assist in general, i can have a convo and everything

#

it's only in this automation and using speak that it seems to fail.

#

could it be that speak service doesn't make the right call via the wyomin protocol to my main server? or just something else i did wrong...

prisma spruce
#

I didn't, i saw some troubles and bailed. Since i used Kokoro on CPU, it was still too slow for me...

gusty eagle
#

Turns out it may be the wyoming protocol middle ware that I’m using. The developer seems to have figured out the issue. I’m using announce instead of speak right now, but will learn to figure out how to update the wyoming-protocol middleware and then update to hopefully also solve speak.

prisma spruce