eden-emu/opus
14
0
Fork 0
mirror of https://github.com/xiph/opus.git synced 2025-05-15 16:08:30 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
opus/dnn/torch/lpcnet
History
Exact
Jean-Marc Valin b0620c0bf9
Using sparse GRUs in DRED decoder 
Saves ~270 kB of weights in the decoder
2023-11-15 04:08:50 -05:00
..
data added copyright headers 2023-09-05 22:31:19 +02:00
engine added copyright headers 2023-09-05 22:31:19 +02:00
models added copyright headers 2023-09-05 22:31:19 +02:00
scripts added copyright headers 2023-09-05 22:31:19 +02:00
utils Using sparse GRUs in DRED decoder 2023-11-15 04:08:50 -05:00
add_dataset_config.py added copyright headers 2023-09-05 22:31:19 +02:00
make_default_setup.py added copyright headers 2023-09-05 22:31:19 +02:00
make_test_config.py added copyright headers 2023-09-05 22:31:19 +02:00
print_lpcnet_complexity.py added copyright headers 2023-09-05 22:31:19 +02:00
README.md added LPCNet torch implementation 2023-09-05 12:29:38 +02:00
test_lpcnet.py added copyright headers 2023-09-05 22:31:19 +02:00
train_lpcnet.py added copyright headers 2023-09-05 22:31:19 +02:00

README.md

LPCNet

Incomplete pytorch implementation of LPCNet

Data preparation

For data preparation use dump_data in github.com/xiph/LPCNet. To turn this into a training dataset, copy data and feature file to a folder and run

python add_dataset_config.py my_dataset_folder

Training

To train a model, create and adjust a setup file, e.g. with

python make_default_setup.py my_setup.yml --path2dataset my_dataset_folder

Then simply run

python train_lpcnet.py my_setup.yml my_output

Inference

Create feature file with dump_data from github.com/xiph/LPCNet. Then run e.g.

python test_lpcnet.py features.f32 my_output/checkpoints/checkpoint_ep_10.pth out.wav

Inference runs on CPU and takes usually between 3 and 20 seconds per generated second of audio, depending on the CPU.