conda create -n hss python=3.7
conda activate hss
git submodule update -i
cd ml_pkg
python setup.py develop
pip install -r requirements.txt
Next step: Move HSS1.5 database into "input" folder.
cd input/
cp -r db1-5/ db15_binary/
rm -r db15_binary/lab
mv db15_binary/binary_lab db15_binary/lab
cp -r db1-5/ db15_3-class/
rm -r db15_3-class/binary_lab
cd "matlabroot/extern/engines/python"
python setup.py install
Ref: https://jp.mathworks.com/help/matlab/matlab_external/install-the-matlab-engine-for-python.html
"ComParE" is the standard feature set for the ComParE challenge, including functionals of 65 low-level descriptors (LLDs) and their deltas, totalling up to 6373 acoustic features. The feature set is extracted with the openSMILE toolkit. The bag-of-audio-words (BoAW) approach is based on the acoustic LLDs, generating a term-frequency histogram representation of quantised audio words. The toolkit openXBOW is used to compute the BoAW features.
Refs are here
Now you setup OpenSMILE.
cd baseline
wget -O opensmile-2-3-0.zip https://www.audeering.com/download/opensmile-2-3-0-zip/?wpdmdl=4781
unzip opensmile-2-3-0.zip
Then, execute this after adjusting "label-dir" and "wav-dir"
python wav4smile.py --input-wav-dir ../input/db15_binary/wav --output-wav-dir ../input/db15_binary/wav_smile
python extractSMILE-XBOW.py --label-dir ../input/db15_binary/lab --wav-dir ../input/db15_binary/wav_smile --task-name HSS1-5_binary
This extracts both the ComParE feature set (6373 descriptors for each instance) and the 130 ComParE LLDs from the wav files of all partitions using openSMILE. Then, openXBOW computes different BoAW representations using codebooks of different sizes. All features are stored in the folder "features/".
- Construct auDeep virtual environment.
cd baseline
git clone https://github.com/auDeep/auDeep.git
conda create -n audeep python=3.5
conda activate audeep
cp ../src/hss15* auDeep/audeep/backend/parsers/
pip install ./auDeep
bash audeep_generate_binary.sh ../input/db_binary/
This is the installation example using conda. Please see other install option here You need to execute those commands after auDeep setup ended.
git clone https://github.com/auDeep/auDeep.git
git submodule update --init --recursive
conda config --add channels pytorch
conda config --add channels conda-forge
conda create -n DeepSpectrum -c deepspectrum deepspectrum
conda activate DeepSpectrum
pip install ./auDeep
bash deepspectrum_generate_binary.sh ../input/db_binary/
Build tensorflow 1.4 with cuda
sudo apt install curl
curl https://bazel.build/bazel-release.pub.gpg | sudo apt-key add -
echo "deb [arch=amd64] https://storage.googleapis.com/bazel-apt stable jdk1.8" | sudo tee /etc/apt/sources.list.d/bazel.list
sudo apt-get update && sudo apt-get install -y bazel
git clone git@github.com:tensorflow/tensorflow.git -b r1.4
git clone https://github.com/end2you/end2you.git