Conversation
The G2P pipeline is direction-symmetric; spell pronunciations, including OOV/made-up ones, through the mirrored chain: - p2g module: P2g trait (ranked candidates), P2gLexicon (exact homophone sets from an inverted FstLexicon), ChainedP2g, ByT5P2g (PolyIPA-style seq2seq, onnx feature) - PhonetisaurusP2g: reversed-WFST reader reusing the same VectorFst parser and Dijkstra search (search over phoneme units, '|'-joined compound inputs) - FstLexicon::from_rows_raw/lookup_raw: exact-case keys for the inverted lexicons (IPA case is significant) - floravox-fst-compile --reverse: phoneme-keyed lexicons with homophones merged per key (first-seen order) - floravox-train-phonetisaurus --reverse: the ilabel/olabel swap (the M2M alignment itself is direction-agnostic); holdout eval reports homophone-aware exact match + CER - floravox-p2g CLI: chain --lexicon/--model/--byt5 tiers Measured on the published en bundle (124k entries, voicegarden- lexicons): 39.2% exact / 41.4% homophone-credited / CER 39.7% / coverage 98.6% — at published Phonetisaurus-P2G baselines; the inverted lexicon tier answers in-lexicon items exactly. No existing G2P code paths change (open() was refactored to a shared table loader; all 48 g2p tests pass unchanged).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The G2P pipeline is direction-symmetric; spell pronunciations,
including OOV/made-up ones, through the mirrored chain:
homophone sets from an inverted FstLexicon), ChainedP2g,
ByT5P2g (PolyIPA-style seq2seq, onnx feature)
VectorFst parser and Dijkstra search (search over phoneme
units, '|'-joined compound inputs)
inverted lexicons (IPA case is significant)
homophones merged per key (first-seen order)
(the M2M alignment itself is direction-agnostic); holdout eval
reports homophone-aware exact match + CER
Measured on the published en bundle (124k entries, voicegarden-
lexicons): 39.2% exact / 41.4% homophone-credited / CER 39.7% /
coverage 98.6% — at published Phonetisaurus-P2G baselines; the
inverted lexicon tier answers in-lexicon items exactly.
No existing G2P code paths change (open() was refactored to a
shared table loader; all 48 g2p tests pass unchanged).