A simple Noise Suppression Model based on RNN and implemented with TensorFlow.
More details: https://aruno14.medium.com/when-your-boss-say-make-noise-suppression-system-8e82ee84afd7
Noisy and Clear data are created using create_noisy.py
.
Models have been test on a very small subdataset (about 20 files). The MSE values are calculated on the training data.
Model | MSE | Size |
---|---|---|
No model | 1.34 | - |
Dense | 1.22 | 854Ko |
Dense_gain | 0.75 | 862Ko |
Dense_image | 0.68 | 367Mo |
Lstm_simple | 1.31 | 8Mo |
Lstm_sequence | 0.56 | 8Mo |
Lstm_image | 0.59 | 5Go |