align-phonemes

Phoneme Forced Aligner

This package was designed to intake human data from Makin Lab (.txt and .wav file from the block needed phoneme alignment) and ouput a JSON file with the mike-on and mike-off times and the alignments (start time, transcription method, production 1, phoneme list 1, production 2, phoneme list 2).

Getting Started

These instructions will give you a copy of the project up and running on your local machine for development and testing purposes.

Installing

This is a pip installable package. Therefore, run the following command:

---

Functions

get_forced_aligment()

input: block txt path, block wav path, output transcription json path, transcription method, critical error threshold, verbose functionality: - determine which transcript will be used based on transcription method input (Critical Error, Wav2Vec, Original) - run Montreal Forced aligner (input trials directory and verbose and returns text grid) - demarcate JSON - clean directories

---

demarcate_to_json()

input: trial directory, block path, text grid directory, output json file path functionality:

• read textgrid
• use ER-demarcation Algorithm to denote phoneme split
• use ER-demarcation to denote transcript split
• write to Phoneme Json (see output format)

---

clean_directories()

input: list of temp directories created functionality: clean and remove directory

---

parse_block()

input: wav file path, txt file path, trials directory, verbose functionality:

• create a Trial Directory (same name as the label): each trial is a .wav and a .txt
• Trial Directory: Trials dir: trial.wav trial.txt ... Trial transcription method .json

TranscriptionMethod.json format: { trial start-time (float): method ('wav2vec' or 'original'), ... }

---

Notes

For a more in depth explanation in the methods used in this package, as well as the reasoning behind, refer back to the paper.

Authors

• James Willian Stonebridge
• Herbert Alexander de Bruyn
• Tyler Dierckman

License

This project is licensed under the MIT License.

Acknowledgments

• Varun implemented the original code for Wav2Vec2 based transcription used in this package.