Files
storyteller-ml/tts/NeMo/scripts/dataset_processing/ljspeech/README
T
2022-02-28 05:43:54 -05:00

15 lines
767 B
Plaintext

These scripts are used for extracting features from LJSpeech.
You will need to run the following scripts (note that these will call the other
scripts as necessary):
- `create_manifests_and_textfiles.py`: Creates the manifest and text files that
are used by the Montreal Forced Aligner (MFA) library.
- `extract_ljspeech_phonemes_and_durs.sh`: Extracts phonemes and alignments via
MFA and calculates phoneme durations.
Needs to be run with Conda active since it creates a Conda environment.
Also note that you may need to install OpenBlas manually (e.g. sudo apt-get
install libopenblas-dev), sometimes MFA can't find it if installed via Conda.
(This script calls `create_token2idx_dict.py` and `calculate_durs.py`.)
- `extract_ljspeech_energy_pitch.py`