Add Thai W2P

[wannaphong]

What does this changes

I add Thai Word-to-Phoneme converter to pythainlp.transliterate.pronunciate. It's converter thai word to thai phoneme.

GitHub: https://github.com/wannaphong/Thai_W2P

Your checklist for this pull request

🚨Please review the guidelines for contributing to this repository.

• Passed code styles and structures
• Passed code linting checks and unit test

[pep8speaks]

Hello @wannaphong! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2021-01-07 13:59:28 UTC

[wannaphong]

Google colab for test https://colab.research.google.com/drive/1MYc5PztkY26-O1bgRXmh9rShx0GLwFWp?usp=sharing

[coveralls]

Coverage increased (+0.03%) to 95.855% when pulling d25fe08 on Add-thai-word2phoneme into 907f2d6 on dev.

[wannaphong]

Model Card

Model Details

• Developer: Wannaphong Phatthiyaphaibun
• Model date: 2020-12-29
• Model version: 0.1
• Used in PyThaiNLP version: __
• GitHub: ___
• License: CC0
• train notebook: https://github.com/wannaphong/Thai_W2P/blob/main/train.ipynb

Intended Use

• Converter thai word to thai phoneme
• Not suitable for other language.

Factors

• Based on thai word to thai phoneme problems.

Metrics

• Evaluation metrics include phoneme error rate (number error / number phonemes)

Training Data

Thai W2P

Evaluation Data

Thai W2P

Quantitative Analyses

epoch: 100
step: 100, loss: 0.03179970383644104
step: 200, loss: 0.04126007482409477
step: 300, loss: 0.01877519115805626
step: 400, loss: 0.03311225399374962
per: 0.0432
per: 0.0419

Ethical Considerations

thai phoneme based on website (wiktionary, Royal Institute et cetera). It may not be the dialect that you use in everyday life.

Caveats and Recommendations

• 1 Thai word only

[bact]

Does it possible now for the transliterate() function to convert multiple Thai words into their sound representation?

If not, I suggest that it should. Because that's how other engine do. We should keep the types of input and output consistent.

This can be done by tokenized the input first, then do the w2p for each token.

[wannaphong]

I changed w2p from pythainlp.transliterate.transliterate to pythainlp.transliterate.pronunciate.

[bact]

I think we're good to go. Thanks @wannaphong

[bact]

All Thai G2P, Thai W2P, etc. can be converted to the singleton pattern but we can do that later.

[wannaphong]

@bact Thank you.