Got all the wyoming services to work also thanks to the solutions in the issue lists also.
I am using a raspberry pi zero w 2 with a respeaker 2 hat
oS = bookworm latetst 64 bit could not get things to work with os bullseye lite 64 bit
wake word works like a charm but speech recognition is very bad and even sometimes things seems to be said that never were said at all which makes no sence.
Any idea how to improve or troubelshoot that .
steps i did was looking in debug under the voice assist trying to set up
assist_pipeline:
debug_recording_dir: /share/assist_pipeline
in homeassistant configuration.yaml but nothing shows up
i tried some settings in the services file but tweaking does not improve recognition
I tried different languages and english seems to be the one working best.
Dutch seems rubbisch.
On both languages the text pipelines work perfect so it seems to the recognition part.
Any help would be appreciated are a link to a working solution would be great too.
This project is unrelated to speech recognition.
My guess is you are using wyoming-whisper for speech recognition.
Try running more advanced models and increasing beam size.
Got all the wyoming services to work also thanks to the solutions in the issue lists also. I am using a raspberry pi zero w 2 with a respeaker 2 hat oS = bookworm latetst 64 bit could not get things to work with os bullseye lite 64 bit wake word works like a charm but speech recognition is very bad and even sometimes things seems to be said that never were said at all which makes no sence. Any idea how to improve or troubelshoot that . steps i did was looking in debug under the voice assist trying to set up assist_pipeline: debug_recording_dir: /share/assist_pipeline in homeassistant configuration.yaml but nothing shows up i tried some settings in the services file but tweaking does not improve recognition I tried different languages and english seems to be the one working best. Dutch seems rubbisch. On both languages the text pipelines work perfect so it seems to the recognition part. Any help would be appreciated are a link to a working solution would be great too.