#Speech to Phrase Processimg Time

1 messages · Page 1 of 1 (latest)

modern estuary
#

Does anyone tried the speech to phrase on a Pi Zero and knows how long the Processing time is compared to Pi4/5. In the blog the time of Pi4 is under a second and the Pi5 at 150 milliseconds. Also if i want to have multiple satellites in House is it the better approach to have StP installed on every satellite or should i stream the audio to a processing endpoint like my homeserver?

limpid yoke
#

I think speech to phrase is meant to be a STT processor that would be part of your voice assistant pipeline in HA and would thus live in HA as an addon, it would not be installed on every satellite.

#

when you wake a satellite it starts streaming the audio to the specified STT from the assist pipeline, then that gets handed to the conversation agent to process, the the pipeline will send the response to TTS and forward that audio response back to the satellite.

modern estuary
#

but speech to phrase can be in its own container? or does it have to be on the HA machine?

limpid yoke
#

I mean technically you could probably run it in its own docker if you are running things that way, not sure there would be any advantage to that other than if you are just running everything as containers instead of HAOS/Supervised install.

modern estuary
#

i prefere this bcs if anything goes bad i dont break my main HAOS system and i can backup my container images

thorn terrace
#

You are aware that addons are Docker containers with a specialized API layer, right? And that you can back them up via HA, locally, to network storage or in the cloud

modern estuary