SungFeng-Huang/libriheavy-mfa

Usage

Step 1: Download LibriHeavy Manifests

• LibriHeavy repo: https://github.com/k2-fsa/libriheavy/tree/master
• Downloaded to:
  • data/LibriHeavy: current repo default
  • or download/librilight: script default
  • or any custom path

Step 2: Load, Cut and Save Audios

Using mfa_librilight.py

Requirements

pip install lhotse textgrid tqdm

Modify path config

• old_prefix = "download/librilight"
  • original prefix in manifests, don't change it
• librilight_dir = "/datasets/LibriLight"
  • change to your LibriLight download path
• libriheavy_dir = "data/LibriHeavy"
  • LibriHeavy manifests download dir, change accordingly
• corpus_dir
  • Directory to place your audio cuts (*.wav) and transcript files (*.lab)

Step 3: Montreal Forced Align

Using scripts/run.sh

MFA environment setup

Installation guide: https://montreal-forced-aligner.readthedocs.io/en/latest/installation.html

I use the docker image option, since the conda installation is slow. Use this repo's .devcontainer/devcontainer.json for changing to root user, or else you can't run the mfa commands successfully.

Modify path & subset configs

Change the subsets to run, corpus_dir, textgrid_dir and cmd_option accordingly.