Malsami

A Dart implementation of a Grapheme-to-Phoneme (G2P) engine for Flutter. This package converts text into phonetic representations that can be used for text-to-speech synthesis.

This library is highly inspired by the Misaki G2P engine.

Currently, only English is supported.

Features

• Convert English text to phonetic representation
• Support for both American and British English pronunciation
• Handle special cases, contractions, and homographs
• Process text with markdown-style phonetic annotations
• Lightweight and easy to integrate with Flutter applications

Getting Started

Add the package to your pubspec.yaml file:

dependencies:
  malsami_dart: ^0.1.0

Import the package in your Dart code:

import 'package:malsami_dart/malsami.dart';

Usage

Basic Usage

// Create an instance of the English G2P engine
final g2p = EnglishG2P();

// Initialize the engine (loads dictionaries)
await g2p.initialize();

// Convert text to phonetic representation
final (phonemes, tokens) = await g2p.convert('Hello world!');

print(phonemes); // Outputs the phonetic representation

Using British English

// Create an instance with British English pronunciation
final g2p = EnglishG2P(british: true);

// Initialize and use as above
await g2p.initialize();
final (phonemes, _) = await g2p.convert('Hello world!');

Using Phonetic Annotations

You can include specific pronunciations for words using markdown-style links:

// The text in the URL part will be used as the phonetic representation
final (phonemes, _) = await g2p.convert('[Kokoro](/kˈOkəɹO/) models');

// Outputs: kˈOkəɹO mˈɑdᵊlz

Example App

Check out the example app in the /example folder for a complete Flutter application demonstrating how to use the Malsami Dart library.

Phoneme Set

Malsami Dart uses a set of phonemes based on the International Phonetic Alphabet (IPA) with some modifications for optimal text-to-speech synthesis. For a complete list of phonemes, see the documentation in the source code.

Limitations

• Currently only supports English
• Requires dictionary files to be included as assets
• No neural network fallback for out-of-vocabulary words yet

Credits

This is a Dart port of the original Python-based Malsami G2P engine, which was designed for Kokoro models.