Just wondered if anyone has created or has any example code of a more recent pytorch using tourchaudio MFCC hopefully in a streaming model that can use Alsa sources?
Or if anybody is up for the idea of kickstarting something.
I quite like the Linto tensorflow HMG (Hotword model generator) as it allows you to create a profile of MFCC & model parameters.
Would be great to have something similar with Pytorch with a GUI where maybe you could test GRU, CRNN, DS-CNN models with datasets.
Yeah without a doubt, more than willing to put in the time.
I was hoping Honk might get an update as a Honk2 and drop Librosa as it can be problematic on some platforms.
But it would be a great addition as when it comes to opensource KWS that model creation is an easy process options are very limited.
Its sort of strange as its easier to find more complex ready to go installs of Kaldi / Deepspeech than a KWS where the models can be easily created from a dataset.
There are quite a few opensource options but often the models are blackboxes and why I posted HMG above as its not just the KWS but model creation tools that would be such a great contribution.
