Wire just released a kind of early alpha for an Open Source Escape Pod on Github. He used the chipper and vector-cloud repositories that were provided by DDL on Github some time ago.
From Wire’s Github repository:
What this is
This is my very-prototype-phase custom escape pod. Right now; I think it has the same STT software, STT model/scorer, and processes text in a very similar way. But, the official escape pod is much much better at intent matching.
chipper is the program which takes Vector’s mic stream (after “hey vector”) and puts it into a speech-to-text processor, then spits out an intent.
This repo has a chipper which has my voice processing code and is from an older tree. The “intent graph” feature seemed to break it a bit.
This also contains a vector-cloud which is also from an older tree and is modified a little bit to allow for a custom cert (instructions somewhere below). This allows Vector to communicate with your custom chipper.
Caveat: This is not for the average user, you will have to compile the code and follow Wire’s instructions. Also this is a first release that still has some problems – those will hopefully be ironed out with the help of other Vector enthusiasts.
Status
Currently, on a fast desktop, the speech-to-text itself is pretty snappy and relatively accurate. But, you have to speak loud and clearly.
Intent matching is barebones right now and only the following are implemented:
go home, start exploring (i recommend saying “deploring”, he understands it better), go to sleep, change your eye color, how old are you, good robot, bad robot
But this definitely is a first step into independency from Digital Dream Labs. They will probably try to suppress this, but since it was derived from the code DDL themselves open sourced I definitely see no illegal activities or copyright infringements here.
But: Fork! Fork! Fork!
This only works with OSKR-unlocked, Dev-unlocked, or Whiskey robots.