Compatible
Matches the phonemes existing Piper, Kokoro, Kitten and Matcha voices were trained on, quirks included, so they keep sounding the way they do.
5× faster than espeak-ng. Just as accurate. Free to ship.
A drop-in phonemiser for Piper, Kokoro and sherpa-onnx voices — English and Hindi, in Rust, Python, C and the browser. Apache-2.0.
12 µs per sentence vs 62 µs · same word error rate through Piper and Kokoro voices · no GPL anywhere
Everything on this page runs locally in WebAssembly; nothing you type leaves your device.
Phonemes
Model input ids
Improved
Fixes what the reference gets wrong: homographs in context, names, numbers and dates, Hindi schwa deletion and nasals. For new voices and for reading text out correctly.
Small and fast
The WebAssembly core is a few hundred kilobytes; each language's data is fetched the first time you use it.
Use it
npm install uccharan
import { phonemize } from "uccharan";
await phonemize("Hello, world."); // "həlˈoʊ, wˈɜːld."
await phonemize("नमस्ते", { lang: "hi", profile: "sherpa-onnx" });pip install uccharan
import uccharan
uccharan.phonemize("Hello, world.") # 'həlˈoʊ, wˈɜːld.'
uccharan.phonemize("नमस्ते", lang="hi", profile="sherpa-onnx")cargo add uccharan
use uccharan::{Lang, Options, Phonemizer};
let p = Phonemizer::new();
p.phonemize("Hello, world.", &Options::new(Lang::EnUs)); // "həlˈoʊ, wˈɜːld."#include "uccharan.h"
UccharanPhonemizer *p = uccharan_new();
char *ipa = uccharan_phonemize(p, "Hello, world.", "en-us", "espeak-ipa", 0);
uccharan_string_free(ipa);
uccharan_free(p);