Open-source notices

Pronunciation and lexical data

The compact pronunciation pack derives from the Carnegie Mellon University Pronouncing Dictionary through the ISC-licensed cmu-pronouncing-dictionarypackage. WordNet indexes provide lemma and part-of-speech metadata, while SUBTLEX-US spoken-frequency counts retain common inflections and improve utility ranking; they are not pronunciation or dialect data. RhymeGraph adds a small, transparent set of slang/reference pronunciations, eight pack fixtures, and 151 ordinary performance-phrase building blocks composed from the word lexicon. None is extracted from lyrics or an artist corpus.

On-device meaning

Semantic retrieval uses sentence-transformers/all-MiniLM-L6-v2, Transformers.js 4.2.0, ONNX Runtime Web, and a RhymeGraph-built int8 index of bounded WordNet gloss/POS documents. WordNet also supplies a bounded primary definition when available. The model, tokenizer, index, runtime, and WASM backend are bundled locally; remote model fallback is disabled.

Application runtime

The static interface includes React, React DOM, Next.js, and Lucide icons. The semantic worker also bundles the MIT-licensed @huggingface/jinjaparser used by Transformers.js. Their copyright notices and licence texts are included with the data, model, and ONNX notices below.

Licence texts