Vector processing text all by himself.
Install Go, gcc, g++, make, automake, git, and wget.
(this can only build on Linux)
make
./deploy.sh <ip>
- This currently works via sherpa-onnx. I chose to convert a Citrinet model to ONNX myself. It's the lowest quality one, but it still works pretty well (better than Vosk in my experience).
- I compiled onnxruntime for Vector's softfp OS with an XNNPACK backend, then compiled sherpa-onnx with that onnxruntime library.
- This was originally using a cut down Vosk model with a limited dictionary. This is now full-fledged speech-to-text running on Vector.
- Not yet, unless you know how to write Go.