hans00/phonemize

Supports multiple output formats (IPA, ARPABET, Zhuyin) and modular multilingual G2P models for English, Chinese, Japanese, Korean, and Russian through preset imports. Uses a processing pipeline combining dictionary lookup (125,000+ entries), intelligent compound word decomposition, and LLM-generated phonetic rules for unknown words, achieving >10,000 words/second with zero startup overhead. Integrates seamlessly into Node.js projects via a simple API with options for custom pronunciations, tokenization with position tracking, and automatic text normalization (numbers, abbreviations, currency, dates).

Used by 1 other package. Available on npm.