- clone the repository
git clone https://github.com/domesticatedviking/TextyMcSpeechy.git
- from the repository directory, run
./install_piper.sh
cd tts_dojo/PRETRAINED_CHECKPOINTS
- from
tts_dojo/PRETRAINED_CHECKPOINTS
rundownload_defaults.sh en-us
to download a set of pretrained checkpoint files. (currentlyen-us
is the only preconfigured language option) - You can use
PRETRAINED_CHECKPOINTS/languages/en-us.conf
as a template for making.conf
files to download piper checkpoints for other languages. Pull requests are welcome. - Copy your audio files and
metadata.csv
file to a new directory inside oftts_dojo/DATASETS
. Keep backups of your original files! - from
tts_dojo/DATASETS
, run./create_dataset.sh <your_dataset_dir>
to set up your dataset. This will sort your files by file format and sampling rate, and automatically create 22050hz and 16000hz.wav
versions of your files if they do not exist. It will also ensure that files mentioned inmetadata.csv
are present. - run
tts_dojo/newdojo.sh <voice_name>
to create a dojo for the voice you are about to build. - inside of
<voice_name>_dojo
, run./run_training.sh
- You will be prompted to choose a dataset, the dataset will be pre-processed, and training will begin.