data.csv

# filename | phonemes | speaker id (speaker id is not used, so anything random is fine)

/path/to/file.wav|hello world but in phonemes|0