PyKokoro Documentation

PyKokoro is a request-centric Kokoro speech-synthesis engine. It accepts prepared speech requests and returns independent rendered waveforms. Parsing, speech planning, document semantics, timeline composition, and final-output mastering belong to the calling application and its other tools.

Contents:

Engine responsibilities

KokoroG2P prepared-text integration with explicit pronunciation languages, source-aligned overrides, and linguistic tokens
Kokoro voice, model, and style selection, including blends
Model token-capacity validation and opt-in sentence-level splitting for oversized requests, plus ONNX inference
Request-local timing reconstruction, waveform validation, tracing, and optional voice calibration
Independent RenderedSegment results and standalone WAV writing

Feature guides

Long-text capacity, splitting, and errors
[Direct phonemes, language routing, VoiceBlend, and result metadata] (advanced_features.md)
English reference voice enrollment and reuse
Language codes and acoustic-profile discovery
Frontends, installed lexicons, cache location, and asset progress
Short-sentence configuration and modes

PyKokoro does not parse SSMD or YAML, create UtterPlans or AudioJobs, compose separate caller requests, insert cross-request silence, or apply document-level effects. The v0.10.0 breaking boundary and migration examples are described in the release note.

Quick example

from pykokoro import GenerationConfig, KokoroSynthesizer, SynthesisConfig

config = SynthesisConfig(
    voice="af_sarah",
    generation=GenerationConfig(lang="en-us"),
)
with KokoroSynthesizer(config) as synthesizer:
    result = synthesizer.synthesize_text("Hello, world.", language="en-us")
result.save_wav("hello.wav")

See the maintained examples for requests with pronunciation context, linguistic tokens, and independent batch results.