I am a ML beginner,I almost finished reading the documents about Flux.jl,but I have no idea about implementing a Bi-directional LSTM using native’s Flux.jl.
I found a tutorial in FluxML/model-zoo.
It is an implementation of the neural network for speech recognition described in Graves & Schmidhuber (2005). It takes in frames of frequency information derived from the waveform, and it predicts which phone class the frame belongs to, among a reduced set of English phones. The training is run using the TIMIT data set (Garofolo et al., 1993).