cookcodes
cookcodes
These codes are from https://github.com/jzi040941/PercepNet, and will be aligned to the changes there.
Yes, Noh revised his codes to avoid out of bound access. But I still have a question regarding the "look ahead", see below: https://github.com/jzi040941/PercepNet/issues/15 I am confused here now.
I user tensorflow 2.3.0 Keras 2.4.3 Yes, GPU is used to train the data. And in rnn_train.py, some numbers need to be changed based on your input data size. traingen...
Before feed the data to neural network, the data are nomalized. It will help to reduce the loss of training result. The mean and variance data are statistic result from...
You are right, it should be standardization instead of normalization.
1 e.g. street_20db.raw in example/ directory, you may use ./rnnnose_demo street_20db.raw street_20db_denoised.raw the output is bad. 2 master branch, use noisyspeech_synthesizer_singleprocess.py to output 48K Hz. data https://github.com/microsoft/DNS-Challenge, interspeech 2021
I do not test the demo on aarch64. Could you check the dump file to see where the crash happen?
For scale bands module, use rb and gb to generate the value in frequency domain. You may check the codes in rnnoise_process_frame() after rb and gb is computed. interp_band_gain(gf, g);...