LPCNet_pytorch

A Pytorch version of LPCNet, including dump weight

• J.-M. Valin, J. Skoglund, A Real-Time Wideband Neural Vocoder at 1.6 kb/s Using LPCNet, Submitted for INTERSPEECH 2019.
• J.-M. Valin, J. Skoglund, LPCNet: Improving Neural Speech Synthesis Through Linear Prediction, Proc. International Conference on Acoustics, Speech and Signal Processing (ICASSP), arXiv:1810.11846, 2019.

Download Data

Use together with the C code of this repo. Suitable training material can be obtained from the McGill University Telecommunications & Signal Processing Laboratory. Download the ISO and extract the 16k-LP7 directory, the src/concat.sh script can be used to generate a headerless file of training samples.

cd 16k-LP7
sh /path/to/concat.sh

Dump Training Data

Use together with this repo.

./dump_data -train input.s16 features.f32 data.u8

Training

python main.py --feat features.f32 --data data.u8

Testing

./dump_data -test test_input.s16 test_features.f32
python test.py --feat test_features.f32

Dump LPCNet weights

python dump_lpcnet.py --load <path/to/checkpoint>