The eSpeak TTS program looks to use a similar algorithm to NRL Report 7948. It's main difference AFAICS is that it uses weights to rank matching rules and picks the one with the best weight instead of using the first matching rule.
Support for this should be provided such that both the plain text format (e.g. from the en_rules file) and the binary file (e.g. en_dict) can be loaded.
This will allow the Cainteoir Engine to use the eSpeak pronunciation rules which are very good at handling unknown words. It will also allow the success rate of those rules to be measured using a master pronunciation dictionary file.
---
Want to back this issue? **[Post a bounty on it!](https://www.bountysource.com/issues/1026788-implement-the-espeak-letter-to-phoneme-algorithm-and-dictionary-files?utm_campaign=plugin&utm_content=tracker%2F254961&utm_medium=issues&utm_source=github)** We accept bounties via [Bountysource](https://www.bountysource.com/?utm_campaign=plugin&utm_content=tracker%2F254961&utm_medium=issues&utm_source=github).
rhdunn
commented
11 years ago
The eSpeak TTS program looks to use a similar algorithm to NRL Report 7948. It's main difference AFAICS is that it uses weights to rank matching rules and picks the one with the best weight instead of using the first matching rule.
Support for this should be provided such that both the plain text format (e.g. from the en_rules file) and the binary file (e.g. en_dict) can be loaded.
This will allow the Cainteoir Engine to use the eSpeak pronunciation rules which are very good at handling unknown words. It will also allow the success rate of those rules to be measured using a master pronunciation dictionary file.