AudioTransform is a package for modifying mono audio files of any sample rate, for use in augmenting datasets of audio.
Operations:
- load/save audio data
- resample
- invert polarity
- scale amplitude
- apply gain adjustments in decibels
- normalize amplitude
- digitally clip amplitude
- pad with silence
- varispeed - adjust time/pitch by altering playback speed
- mix two audio files
- static low-pass filter
- dynamic low-pass filter
- static high-pass filter
- convolution reverb
# make virtual environment with Python 3.7
conda create -n AudioTransform python=3.7
source activate AudioTransform
# clone repository
git clone [email protected]:speechLabBcCuny/AudioTransform.git
cd AudioTransform
# install requirements from pip
pip install -r requirements.txt
# set up jupyter kernel
conda install ipykernel
ipython kernel install --user --name=AudioTransform
# start Jupyter server
jupyter notebook --no-browser --port=8889
# on local machine
ssh -N -f -L localhost:8888:localhost:8889 [user]@[server]
See AugmentAudio.ipynb for examples of how to process audio.
Note - transformations are applied to the object itself, allowing method chaining for complex signal chains.
If a transformation is only being applied temporarily (e.g. scaling amplitude during mixing), use the copy() method to create a new audio file to transform.
Example:
audio1 = AudioFile('example1.wav')
audio2 = AudioFile('example2.wav')
# audio1 is mixed with (audio2 * 0.5) - after this operation, audio2 is now scaled by 0.5
audio1.mix(audio2.scale(0.5), relative_start=0.0)
# The following line will mix audio1 with (audio2 * 0.5) - audio2 remains unchanged by this operation
audio1.mix(audio2.copy().scale(0.5), relative_start=0.0)
python3 <recipe_name>.py -i <input_dir> -o <output_dir>