alphacep--vosk-api
This is still work in progress, more to come
TODO
-
Optimize graph construction, current one is below accuracy
-
Load model from the AAR (mmap them in tflite style)
-
Add decoding speed measurement
-
Add wakeup word
-
Add speakerid
-
Integrate proper hardware optimized neural network library. Candidates are:
-
Quantization for the models