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.
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.
./concat.sh /scratch2/mlavechin/GIPSA/training_sets/lpcnet/archives input0.s16
./dump_data -train input0.s16 features0.f32 data0.u8
python main.py --feat features0.f32 --data data0.u8
./concat.sh /scratch2/mlavechin/GIPSA/training_sets/pb2009/blabla input1.16
./dump_data -train input1.s16 features1.f32 data1.u8
python main.py --feat features1.f32 --data data1.u8
Use together with this repo.
./dump_data -train input.s16 features.f32 data.u8
python main.py --feat features.f32 --data data.u8
./dump_data -test test_input.s16 test_features.f32
python test.py --feat test_features.f32
python dump_lpcnet.py --load <path/to/checkpoint>