English-G2P-Lexicon
An American English pronunciation dictionary for grapheme-to-phoneme conversion in a web browser, as used by Quick Inference's text-to-speech models that read phonemes (Kitten TTS, Inflect, Paradee). It needs no eSpeak.
Contents
en-us-lexicon.txt.gz (1.0 MB): the us_gold and us_silver lexicons of Misaki
(Kokoro's G2P), compacted. Words are front-coded, phoneme symbols recoded to ASCII and heteronyms tagged with their
part of speech. Entries that Misaki's own rules rebuild exactly (plurals, past tenses and so on) are left out
(183,561 → 152,816 entries); the reader derives them as Misaki does.
Changes from the original
Only the format: the pronunciations are Misaki's. Built by tools/export/build_g2p_lexicon.py in
Quick Inference.
Licence
Apache 2.0, as Misaki (LICENSE-misaki.txt, with a note of the changes).