breizhn / DTLN

Tensorflow 2.x implementation of the DTLN real time speech denoising model. With TF-lite, ONNX and real-time audio processing support.
MIT License
567 stars 160 forks source link

about traing a real time model #36

Closed AaZz101 closed 3 years ago

AaZz101 commented 3 years ago

Hello, in your code about training, you used 15s of data and then divided into frames, so the timestep of the data input to lstm is the number of frames, then I want to know about if I want to train a real-time processing model, Then the input data during training is single frame data, like (batch,timestep=1,512) or 15s data like (batch,timestep=1873,512)