new-fave
What is new-fave?
new-fave is a tool for automating and optimizing vowel formant extraction. It is philosophically similar (and named after) the FAVE-suite. However, new-fave has been completely written from scratch, and has some key differences from the FAVE-suite.
new-favedoes not include a forced-aligner. It can process alignments produced by fave-align, but we would recommend using the Monteal Forced Aligner insteadnew-favedoes not require speaker demographics. You can optionally passfave-extracta speaker demographics file to be merged into your formant data, but this does not influence how the data is processed in any way. Besides including file name and speaker number data, you can pass any demographic information you would like.new-favedoes not assume North American English vowels. Your alignments can contain any set of vowels, in any transcription system, as long as you can provide a regular expression to identify them.new-faveis customizable. With config files, you can customize vowel recoding, labelset parsing, and point measurement heuristics.new-faveis focused on formant tracks. You can still produce single point measurements for vowels, butnew-faveis built upon the FastTrack method. By default, it will write output files including point measurements, full formant tracks, and Discrete Cosine Transform coefficients.new-faveis maintainable. As time goes on, and the code base needs updating, the organization and infrastructure ofnew-faveshould allow it to be readilly updateable.
You can read more on the getting started page.
Installation
You can install new-fave with pip.
# bash
pip install new-faveUsage
To use the default settings (which assume CMU dictionary transcriptions), you can use one of these patterns.
A single audio + textgrid pair
# bash
fave-extract audio-textgrid speaker1.wav speaker1.TextGridA directory of audio + textgrid pairs
# bash
fave-extract corpus speakers/Multiple subdirectories of audio + textgrid pairs
# bash
fave-extract subcorpora data/*