Adapted from facebookresearch/fairseq

git clone --recursive https://github.com/seanghay/khmer-forced-aligner.git

Inference

Resample the audio to be 16kHz

ffmpeg -i input_audio.ogg -ar 16000 -ac 1 -c:a pcm_s16le audio.wav

Align

python align_and_segment.py \
  -a audio.wav \
  -t audio.txt \
  --lang khm \
  --outdir results/audio \
  --uroman_path uroman/bin