Using FAVE-Align on Phonetics Lab Mac

The following description is meant to expand on the information included in https://github.com/JoFrhwld/FAVE/wiki/Using-FAVE-align which is the actual guide to using FAVE Forced Alignment. You should absolutely go read it first and come back to these instructions for help running forced alignment in our lab.

To better show how to use FAVE, I am going to include specific examples based on a sound file called Joy-2019-4-23.wav in a folder called ~/Documents/Example. In the specific directions or commands like those in the second list under 3, you’ll want to change those to fit your project.

  1. To start with, you need your sound file and a transcription of what is being said. Your transcription needs to match the audio pretty well so you can either edit the transcription to match what the speaker says in the audio or trim speech you don’t want out of the recording in praat. You can listen to Joy-2019-4-23.wav in the Example folder under Documents on the phonetics lab mac for an example.

  2. The audio file will be your first input into FAVE, a tab-delimited (which means that pressing tab on the keyboard will make a new column) text file with five different columns. As you read in the instructions, the first column is a brief code for the speaker, like a participant number. The second column is the speaker’s full name or pseudonym, but if it is an anonymous study you can list the participant number again here. The third column is the onset time, in seconds (usually 0 seconds unless you are starting in the middle of an audio file, if so you can get this number from Praat). The fourth column is the offset time, in seconds (which is usually the end of the audio file or where yo would like FAVE to stop, either way you can get this number from Praat). The last column is the transcription of the speech between the onset and offset times you just made sure matched the audio in the last step. You can find my tab-delimited file called Joy-2019-4-23.txt in the Example folder under Documents on the phonetics lab mac for an example.

  3. Before FAVE can actually align the trascription to the audio, you have to make sure it knows the phonetic transcription to all of the words. To do this:

    1. Open the Terminal on the Mac and run cd /Users/phonetics/Documents/FAVE/FAVE-align which puts us in the FAVE directory.
    2. Then check for unknown words by running ./FAAValign.py -v -c unknown.txt /Users/phonetics/Documents/Example/Joy-2019-4-23.txt
    3. After you have dealt with the unknown words, run forced alignment with python FAAValign.py -v -i /Users/phonetics/Documents/Example/input.txt /Users/phonetics/Documents/Example/Joy-2019-4-23.wav /Users/phonetics/Documents/Example/Joy-2019-4-23.txt
    4. Finally, you need to open each wav file/TextGrid pair and check that the aligner worked as you expected (often you will need to fix errors by hand).

  4. Open the .TextGrid and .wav files in Praat and make any adjustments to the alignment necessary. You can also open Joy-2019-4-23.wav and Joy-2019-4-23.TextGrid for a pretty good example of alignment.