inflectg2p

An Inflect v2 phoneme and token-ID frontend for prepared, speakable English text.

inflectg2p sends text through espeakng-runtime, applies the Inflect symbol inventory, and intersperses VITS-style blank IDs. It does not expand numbers, dates, times, currency, abbreviations, identifiers, or other semantic written forms. Prepare those in your application before calling this package. It does not load a TTS model or synthesize audio.

Install

python -m pip install inflectg2p

Optional runtime and lexicon extras:

python -m pip install 'inflectg2p[bundled]'
python -m pip install 'inflectg2p[lexphon]'
python -m pip install 'inflectg2p[g2lex]'

Lexicon packages and pronunciation assets are optional. inflectg2p never downloads lexicons implicitly. See installation and lexicons.

Python

from inflectg2p import InflectG2P

with InflectG2P() as g2p:
    result = g2p.phonemize_prepared("Hello, world.")

print(result.phonemes)
print(result.token_ids)
print(result.token_count)

result.text is the unchanged input. result.lexicon_hits reports explicit lexicon matches when configured. The functional helpers phonemize_prepared, phonemes, and phoneme_ids are also available.

CLI

inflectg2p 'Hello, world.'
inflectg2p --espeak-mode cli 'Hello, world.'
inflectg2p --lexicon en-us:lexhint 'Saskatchewan is fluorescent.'

The CLI prints JSON containing the prepared text, phonemes, token IDs, and token count. There is no --normalize option because semantic preparation belongs to the caller.

Documentation and examples

Start with the documentation index. Runnable examples are in examples/.

Compatibility

The exact contract covers the Inflect v2 symbol inventory, strict symbol validation, and blank framing. eSpeak pronunciation depends on the installed runtime and voice data. See compatibility for the boundaries and golden snapshots.

Version and license

Versions are derived from Git tags with setuptools-scm. The project is licensed under Apache-2.0; see LICENSE and NOTICE.