Need help with speech-denoising-wavenet?
Click the “chat” button below for chat support from the developer who created it, or find similar developers for support.

About the developer

520 Stars 149 Forks MIT License 3 Commits 29 Opened issues


A neural network for end-to-end speech denoising

Services available


Need anything else?

Contributors list

# 57,503
3 commits

A Wavenet For Speech Denoising

A neural network for end-to-end speech denoising, as described in: "A Wavenet For Speech Denoising"

Listen to denoised samples under varying noise conditions and SNRs here


It is recommended to use a virtual environment

  1. git clone
  2. pip install -r requirements.txt
  3. Install pygpu

Currently the project requires *Keras 1.2** and Theano 0.9.0, the large dilations present in the architecture are not supported by the current version of Tensorflow (1.2.0)*


A pre-trained model (best-performing model described in the paper) can be found in

and is ready to be used out-of-the-box. The parameterization of this model is specified in

Download the dataset as described below



THEANO_FLAGS=optimizer=fast_compile,device=gpu python --mode inference --config sessions/001/config.json --noisy_input_path data/NSDTSEA/noisy_testset_wav --clean_input_path data/NSDTSEA/clean_testset_wav

To achieve faster denoising, one can increase the target-field length by use of the optional

argument. This defines the amount of samples that are denoised in a single forward propagation, saving redundant calculations. In the following example, it is increased 10x that of when the model was trained, the batch_size is reduced to 4.

Faster Example:

THEANO_FLAGS=device=gpu python --mode inference --target_field_length 16001 --batch_size 4 --config sessions/001/config.json --noisy_input_path data/NSDTSEA/noisy_testset_wav --clean_input_path data/NSDTSEA/clean_testset_wav


THEANO_FLAGS=device=gpu python --mode training --config config.json


A detailed description of all configurable parameters can be found in

Optional command-line arguments:


Valid Inputs Default Description
mode [training, inference] training
config string config.json Path to JSON-formatted config file
printmodelsummary bool False Prints verbose summary of the model
load_checkpoint string None Path to hdf5 file containing a snapshot of model weights

Additional arguments during inference:


Valid Inputs Default Description
oneshot bool False Denoises each audio file in a single forward propagation
targetfieldlength int as defined in config.json Overrides parameter in config.json for denoising with different target-field lengths than used in training
batchsize int as defined in config.json # of samples per batch
conditionvalue int 1 Corresponds to speaker identity
cleaninput_path string None If supplied, SNRs of denoised samples are computed


The "Noisy speech database for training speech enhancement algorithms and TTS models" (NSDTSEA) is used for training the model. It is provided by the University of Edinburgh, School of Informatics, Centre for Speech Technology Research (CSTR).

  1. Download here
  2. Extract to

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.