kitteng2p

Independent phoneme and token-ID frontend targeting the public KittenTTS v0.8 text-cleaner contract.

kitteng2p turns prepared, speakable text into eSpeak phonemes, Kitten-prepared phoneme text, character IDs, and model framing (0, *ids, 10, 0). It does not synthesize audio, load ONNX models, download voices, or perform written-to-spoken semantic normalization. Callers prepare dates, currencies, units, URLs, and abbreviations before phonemization.

The package has a flat layout and uses setuptools-scm for dynamic versioning.

Installation

Install kitteng2p and make a system eSpeak or eSpeak NG runtime available to espeakng-runtime:

python -m pip install kitteng2p

Where supported, install the optional bundled runtime assets instead:

python -m pip install "kitteng2p[bundled]"

The bundled extra provisions eSpeak runtime assets where supported. It does not install KittenTTS, a TTS model, or audio dependencies.

Quick start

from kitteng2p import KittenG2P

with KittenG2P() as g2p:
    result = g2p.phonemize_prepared("Hello, world.")

print(result.phonemes)
print(result.token_ids)
print(result.dropped_symbols)

The result's token_ids already include model framing. The convenience functions phonemes() and phoneme_ids() create and close a temporary frontend.

Command line

kitteng2p "Hello, world."
kitteng2p --language en-gb "Hello, world."
kitteng2p --espeak-mode cli "Hello, world."

The command prints one JSON object with the original text, prepared phonemes, framed token IDs, and any dropped symbols.

Kitten v0.8 compatibility target

kitteng2p reproduces the public KittenTTS v0.8 text-cleaner symbol inventory and model-ID framing:

• the same pad, punctuation, ASCII-letter, and IPA symbol inventory;
• duplicate symbols use the last dictionary index, matching upstream construction;
• symbols outside the inventory are dropped;
• model IDs are framed as (0, *ids, 10, 0).

This is a codec-level compatibility claim. The released KittenTTS 0.8.1 path uses phonemizer around eSpeak, while kitteng2p uses espeakng-runtime. Exact sentence-level ID parity has not been established across a representative corpus. Do not interpret codec compatibility as end-to-end phonemizer parity. See the compatibility policy.

Documentation

• Installation
• Quick start
• API reference
• Codec contract
• eSpeak backend
• Command line
• Compatibility policy
• Architecture
• Changelog
• Runnable examples

Dynamic versioning

Package versions are derived from Git tags through setuptools-scm. Source archives without Git metadata use the configured 0.1.dev0 fallback. Create a release tag only after the release checklist and artifact gates have passed. At runtime, kitteng2p.__version__ reports the installed distribution version.

Development

python -m pip install -e ".[dev,docs]"
python -m pytest
python -m ruff check kitteng2p tests examples
python -m mypy kitteng2p
python -m compileall -q kitteng2p tests tools examples
python docs/make.py html
python -m build --sdist --wheel