Bidirectional CPC Model for Speech Representations

The bidirectional CPC model proposed in the paper:
"On Scaling Contrastive Representations for Low-Resource Speech Recognition".

See example.py for a simple example of how to load the model and extract the representations used in the paper.

File format used for training was 16-bit PCM sampled at 16kHz. See test.wav.

To test, run:

conda create -n scaling python==3.6.8
conda activate scaling
pip install -r requirements.txt
python example.py

Tested with CUDA 9.0.