Cross-platform mobile Neural Network library for training and inference on device. CPU only.
On Apple platforms it uses Accelerate framework as a backend; On other platforms it uses Eigen for matmul, kissfft for dft, and ARM NEON intrinsics for acceleration.
NN layers: Conv1d GRU RNN LSTM BatchNorm Activation TimeDistributedDense Dense
DSP tools: Spectrogram Window functions
Algorithms were tested using tf 2.3.0
Examples and tests will be exposed in different repository.