English LTS

English letter to sound tool

cmudict

• http://www.speech.cs.cmu.edu/cgi-bin/cmudict

cmudict with syllables

• http://webdocs.cs.ualberta.ca/~kondrak/cmudict.html
• https://en.wikipedia.org/wiki/Syllabification

wagon CART tree

• http://www.cstr.ed.ac.uk/projects/speech_tools/manual-1.2.0/x3475.htm

Grapheme-Phoneme Alignment

This project aims to implement a algorithm proposed in paper as follows:

Pagel V, Lenzo K, Black A. Letter to Sound Rules for Accented Lexicon Compression[J]. 1998, cmp-lg/9808010.

Project Structure

• g2p_data_pre_process.py - a python script to pre-process g2p data by rules.
• g2p_aligner.py - a python script to train model by EM & DTW algorithm.
• assets/ - store the training data sets.
• log/ - settle the log files.

G2P Data Preprocess

Try to match every word-phones pair to every rule based on the locations of grapheme & phoneme.

how to match word-phones pair to a rule.

1. Match a word-phones pair to a rule based on the locations of grapheme & phoneme.
2. If a rule got matched, modify the phones and rematch the word-phones pair to every rule, until no rule would match.
3. Set a admissible error to admit some error caused by previous silent graphemes(letters).