NoiseReductionUsingGRU

Graduation Project Title: Noise Reduction Using GRU(RNN).

This is my graduation project in BIT, 2018.

The project based on Python 3.5 and TensorFlow 1.8.
The project learning from Noise Reduction using RNNs with Tensorflow by adityatb.
Thanks for adityatb's work! adityatb uses LSTM to build noise reduction model.

With the development of RNN, I think GRU may be a better method.
This graduation project uses MIR-1K dataset and try to use GRU to build noise reduction model.

Introduction

This project includes 2 main part.

1. CreateNoiseAddedDataset
   This script will add 3 kinds of noises(Brownian, Pink and White) to the clean human voice.
   If you have 1000 human voice files(MIR-1K has 1000 usable files), you will get 3000 noise added files.

2. GRUTraining
   This is a big part including 5 scripts.

   1. CreateTestDataset.py
      In last step, we have got 3000 files.
      By this script, we randomly separate 1000 files to be used in Test process.

   2. LSTMTestTraining.py
      This script and the next script aim to check LSTM model's performance.
      To see the GRU model works or not.

      adityatb used LSTM model, but the original code has a lot of problems:
      >Python 2.x -> Python 3.x
      >Array overflow problems
      >Fourier trasformation using error
      >No Learning Rate auto change
      >...
      I solved a lot. It is hard to introduce all at here.
      Please refer to the history of git publishment.

   3. ModelTest.py
      This script will use the LSTM model trained in last script and execute 1000 test sounds.
      To listen and compare the spectrum of [clear voice], [noise added sound] and [output] to see the performance.

   4. GRUTraining.py
      This script is modified by LSTMTestTraining.py.
      Change LSTM to GRU in TensorFlow framwork is very easy.

   5. GRUModelTest.py
      This script is modified by LSTMTestTraining.py/

Execute Steps

NOTICE:The detail of the implement, please read the code, refer to the modification history.

1. CreateNoiseAddedDataset
   Download MIR-1K dataset. Decompress the folder Wavfile. We will get 1000 wav files.
   Set all wav files into the Wavs folder.
   Run CreateNoiseAddedDataset.py.
   We will get 1000 human voice files in ./Training/HumanVoices.
   3000 noise added sounds in ./Training/NoiseAdded.

2. GRUTraining

   1. CreateTestDataset
      Cut the ./Training folder into this part folder.
      Run CreateNoiseAddedDataset.py.
      We will get 1000 random chosen noise added files in ./Testing/NoiseAdded.
      And in ./Training/NoiseAdded will 2000 files left.

   2. GRUTraining
      Run GRUTraining.py.
      This script will train a GRU model from remaining 2000 files and corresponding pure human voice files.
      Finally we will get TensorFlow checkpoint files in ./TFCheckpoints folder.

   3. GRUModelTest
      Run GRUModelTest.py.
      We will get less than 1000 files(because some noise added files correspond to the same pure human voice file) in ./Testing/ModelOutput folder.
      You can test the model or do something else you like.

FINISH!