Linguistic-Data-Consortium / p2fa

code for working with the Penn Phonetics Lab forced aligner

p2fa

Code for working with the Penn Phonetics Lab Forced Aligner (p2fa)

Introduction

The Penn Phonetics Lab Forced Aligner is still available from the Lab website:

https://web.sas.upenn.edu/phonetics-lab/facilities/

From its documentation:

P2FA can be cited as:
Jiahong Yuan and Mark Liberman. 2008. Speaker identification on the SCOTUS corpus. Proceedings of Acoustics '08.

For full documentation, see that package. This repo contains the primary components of p2fa and additional code for working with it. The following checksum file documents the exact zip file from that website that was the basis for this repo:

sums/Penn-Phonetics-forced-aligner-2d2jfwb.zip.md5

Model Directory

The "model" directory, copied here from the Phonetics Lab release, contains the model files for English, as well as a version of the CMU pronouncing dictionary, documented with this checksum file:

sums/model.md5

HTK

HTK version 3.4.1 must be acquired separately. The docker recipes assume the presence of the unpacked "htk" directory.

Docker Examples

This repo is designed to be used with docker, and contains several examples. Each example contains a Dockerfile and a readme with example commands, which are meant to be run from the current directory, not the example directories. If you prefer not to use docker, the recipes serve as documentation for configuring your environment.