espeak-ng / espeak-ng

eSpeak NG is an open source speech synthesizer that supports more than hundred languages and accents.
GNU General Public License v3.0
4.04k stars 865 forks source link

Support more dialects of English #190

Open cmrdt opened 7 years ago

cmrdt commented 7 years ago

instead to phoneme substutions maybe try % this & % that region's vowels, question of whether to completely vowel drop distinctions (merger)

and maybe add support for american english variants https://en.wikipedia.org/wiki/North_American_English_regional_phonology https://en.wikipedia.org/wiki/North_American_English https://en.wikipedia.org/wiki/Inland_Northern_American_English https://en.wikipedia.org/wiki/California_English or https://en.wikipedia.org/wiki/Australian_English

rhdunn commented 7 years ago

The English voices use custom phonemes like A@ that map to John Wells' Lexical Sets comparing RP to GenAm, with an extension to support the unmerged Scottish NURSE vowel (FIR-FUR-FERN). These could be extended to add new phonemes that support different dialects of American English, or English in other countries.

Using this approach, with the merges described in the voice (language/dialect) files, makes the phoneme rules more maintainable.

NOTE: The Scottish FIR and FERN vowels should be done in the same way, allowing for Irish English support.

nmstoker commented 5 years ago

It has been quite a while since this was raised - has there been any further thought or development relating to this?

I think it would be really useful to tackle supporting a wider range of dialects.

Recently I was looking through the en-GB-x-rp output and noticed a few inconsistencies that I don't consider to be a traditional RP way of saying words (eg the g for exit rather than an x) and I'd be happy to start contributing proposed changes but I didn't want to do that a) if it was going to break things for more general users and b) if the way things are stored is likely to change, as it would be better to make the contributions after any new approach is agreed.