/viet-aligner

Aligner vietnamese text and audio clip

Primary LanguagePythonMIT LicenseMIT

viet-aligner

Aligner vietnamese text and audio clip

sudo apt install -y sox libsox-fmt-mp3

Install MFA

conda create -n kaldi python=3.9.12
conda activate kaldi
conda install -c conda-forge kaldi montreal-forced-aligner

Download datasets

python download_infore.py --output-dir data
python download_vivos.py --output-dir data
python download_common_voice.py --output-dir data
python download_fpt_open_speech.py --output-dir data

Align using pretrained acoustic model

mfa align /path/to/corpus/dir assets/lexicon.txt assets/mfa_vi_model.zip /path/to/output/dir

Train new acoustic model

python create_lexicon.py --data-dir data
mfa train --clean -o mfa_vi_model -t ./mfa_tmp ./data ./data/lexicon.txt mfa_output