Dataset Open Access
LA-2A Compressor data to accompany the paper "SignalTrain: Profiling Audio Compressors with Deep Neural Networks," https://arxiv.org/abs/1905.11928
Accompanying computer code: https://github.com/drscotthawley/signaltrain
A collection of recorded data from an analog Teletronix LA-2A opto-electronic compressor, for various settings of the Peak Reduction knob. Other knobs were kept constant.
Audio samples present in these files are either 'randomly generated', or downloaded audio clips with Create Commons licenses, or are property of Scott Hawley freely distributed as part of this dataset.
Data taken by Ben Colburn, supervised by Scott Hawley
Revisions in v.1.1:
Made the following corrections to discrepancies in original dataset:
Only one of file: 235, 236
$ rm Train/target_235_LA2A_2c__0__70.wav
$ rm Val/input_236_.wav
In wrong directory: 245
$ mv Train/input_245_.wav Val/
Mismatched length and time alignment: 148, 148, 149, 150, 152
All were had targets delayed by 8583 samples relative to inputs, and were shorter.
Truncated beginning of inputs to make them the same as targets. Used new script check_dataset.py to fix & overwrite earlier version:
$ signaltrain/utils/check_dataset.py --fix SignalTrain_LA2A_Dataset/