Custom host for Whisper STT

[icelaglace]

Hello,
I did a tiny change on Wire-pod to support custom Whisper endpoints.

Why ? :

• My Pi is too weak for anything else
• I don't have an OpenAI subscription
• I'm already running a 24/7 Whisper server on another machine

I had to change the STT silence period to 1 second as it feels much nicer to me, it used to cut on my while I was talking. 😂

2 changes that I'm asking for help with :

• Why is there 2 error checks at same place doing the same thing? I modified the last one to catch EOF which happens if the Whisper request is super long like >10 seconds.
• I don't see any conversion to mp3 anywhere and on my local server, WAV just worked and is direct. Is it just sending a PCM to WAV renamed as MP3 ?

Thanks! :)

[kercre123]

The whisper (non-cpp) implementation was experimental, and a result of copying a set of functions from one project to another project and quickly adapting it; thus the mess.

I like what you have done, though the inactiveNumMax is a big change which affects every STT service. I'd accept a slight increase. 40 or so would be fine.

[icelaglace]

True that, my bad. It might affect all users even if they don't use Whisper. I reverted back the change, sorry.
What do you think of making it editable by the user via either env or web UI at some point ?
I get why you selected 23 though, it's very snappy for built-in intents :)
Thanks!