I just finished setting up my internal voice assistant, but am having troubles getting responses in reasonable time. I’m currently using HA to link the STT, NLP and TTS installations together, I’m using the llama3.1:8b model and exposed no HA entities to the assisstant. Debug shows STT 0.8s, NLP 0.16s and TTS 0.0s, yet end to end it takes 13secs to hear the response. I understood that this is 100% down to HA -by name the pipeline orchestration. In comparison to cloud based solutions the voice preview device waits a few seconds after the last word was spoken and does not stream during saying the sentence / predict the end of the sentence.
I am now wondering if i should take the voice assistant completely out of HA and use HA only as the home control backend. Not sure if i can still just say turn lights on, etc, but i got the understanding this is currently the only way to get to a Amazon echo like experience. Does anyone have any experience/other suggestions?
The debug offers me only to run the text pipeline which won’t show the delay. The two voice options are greyed out - apparently by design as HA has no insight into what the voice preview device does.
I think you misunderstood me. I know where to find debug, but it doesn’t capture the timings within the voice assistant preview device as mentioned before.