Open rugene76 opened 4 months ago
I also had this problem but had to change it to "gpt-4o-mini-2024-07-18". Try that!
how the heck did you figure that out? !! That worked! yea! On to the next issue. btw, I've put together both an esp32 and raspi w2 AI hardware, after much tweaking none of them are working well enough to replace my google hubs. Too slow to respond, unreliable wakeword, false wakeword detections....frustrating.
@rugene76 try procupine1 instead of open-wakeword. For me that made all the difference.
I have working procupine3 in my repo (is modded to work with rk3588) https://github.com/Al3XKOoL/hassio-addons
Its working for me and custom wake words are working. Advantages of v3 compared to v1:
Improved accuracy More supported languages and pre-trained wake words Ability to use custom wake words Drawbacks:
An access token from console.picovoice.ai is needed Each token can be used on only 3 machines per month
For stt i use vosk. With large model heat a lot of ram but its really faster. https://github.com/rhasspy/hassio-addons
I can't get the integration to switch to gpt-4o-mini model.
I enter "gpt-4o-mini" into the field in the configure dialog, reload and even reboot HA but the chatgpt reports that its using the 3.5 architecture.
What am I missing?
thanks