You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
There is currently no G2P (grapheme-to-phoneme) support for Georgian in NeMo's TTS pipeline. This makes it impossible to train or run inference on Georgian text using NeMo's TTS models without building a custom G2P solution outside the framework.
Describe the solution you'd like
Add a "ka-GE" locale to the existing G2P infrastructure:
Register Georgian grapheme and IPA character sets in ipa_lexicon.py
Add a GeorgianG2p class with a rule-based character-to-IPA mapping table
Georgian has a perfectly phonetic orthography — every letter maps to exactly one phoneme with no exceptions, so a simple mapping table is sufficient as the core G2P logic
Adding a dictionary-based lookup (like other locales use), but Georgian's perfectly phonetic script makes this unnecessary overhead. A dictionary could be added later for edge cases if needed.
Additional context
Georgian uses the Mkhedruli script (33 letters, Unicode U+10D0–U+10FF)
No uppercase/lowercase distinction in modern usage
~3.7M native speakers, official language of Georgia
IPA phoneme inventory is well-documented in linguistic literature
eSpeakNG already supports Georgian, confirming IPA mappings are standardized
Similar in scope to Add Urdu (ur) G2P support to TTS pipeline #15445 (Urdu G2P)
I have prior experience building Georgian TTS pipelines using NeMo magpie tts and fastpitch and am willing to contribute a PR for this.
There is currently no G2P (grapheme-to-phoneme) support for Georgian in NeMo's TTS pipeline. This makes it impossible to train or run inference on Georgian text using NeMo's TTS models without building a custom G2P solution outside the framework.
Describe the solution you'd like
Add a "ka-GE" locale to the existing G2P infrastructure:
Usage would follow the existing IpaG2p pattern:
Describe alternatives you've considered
Adding a dictionary-based lookup (like other locales use), but Georgian's perfectly phonetic script makes this unnecessary overhead. A dictionary could be added later for edge cases if needed.
Additional context
I have prior experience building Georgian TTS pipelines using NeMo magpie tts and fastpitch and am willing to contribute a PR for this.