Semester project for Sound and Music Computing (AAU SMC 2018)
Link to report: https://www.overleaf.com/project/5c94e7edf54b1f21e4c1c066
Link to datasets: https://drive.google.com/drive/u/0/folders/1MmtiRQF-33Gkx1ZJZ7vSDCzy-vqjaUw6/
Implementation of a single-channel denoising system in voice detection applications. The system would be based on deep learning techniques (e.g. autoencoders).
Initially, a baseline model shall be chosen, and an evaluation procedure established. Subsequently, improvements comprising techniques drawn from relevant literature will be implemented into the baseline, and evaluated accordingly. The project will be carried out in an iterative fashion.
Should satisfactory performances be achieved within a subset of the timeframe, the following further developments could be considered:
conda create --name <env_name> --file spec-file.txt
source activate <env_name>
spec-file.txt
is modified, update environment:
conda install --name <env_name> --file spec-file.txt
Create new model as in models/
as model_<model_name>.py
.
get_model()
method which returns a keras.Model
objectget_lossfunc()
method with return a loss function taking x_pred
and x_true
as argumentsmodel_example.py
for referencepython main.py --help
main.py
: scripts entry pointscripts/
: scripts for training a model, viewing results, and using encoder and decoderlibs/
: code dependencies for scriptsmodels/
: model architecture implementationsnotebooks/
: jupyter notebooks for experiments and teststools/
: miscellaneous software toolsPipfile
, Pipfile.lock
, environment.yml
: list of dependencies, used for setting up pipenv (local) and conda (remote) environments
notes/
: minutes from group meetingsliterature/
: relevant papers sorted by categoryext/
: unsorted, mixed stuff