Open leibny opened 4 years ago
I know, I know, you are thinking about the Tensorflow Playground! No, this is not in the plans.
While general 4GL never succeeded, there were successful tools to construct dataflows and even the very neural networks. However, they've never achieved a huge success. There is usually more nuances to the design of a NN solution that seems at the first sight (I lived through the times when people were all gung-ho about 4GL, and thought of 5GL—away with Fortran, welcome… eh, sorry, I forgot. C and Fortran are still around, tho)
A well thought-through contribution would be certainly welcome, however. Maybe even better: it may be a product n itself, supporting Kaldi as one of its many targets.
But again, to reiterate, no, we're not planning any graphical building tools at the moment.
Many thanks to Kaldi developers and Dan Povey for writing such a great toolkit. I have a couple of questions ... -- Q1: Is it useful to write a toolkit which allows for building a Kaldi-based ASR system graphically, sth like Matlab simulink by drag and drop pre-defined configurable objects?