This is SpecAugment code with numpy and scipy.
In my test environment, this code is about 100 time faster than using tensorflow.
pip install -r requirements.txt
python3 specaugment.py
if you want to test your test audio, run this command
python3 specaugment.py -i <input file>
start to SpecAugment 100 times
whole processing time : 0.9117 second
average processing time : 9.12 ms
- original spectrogram
- warped spectrum
- warped + masked spectrum
