G2P Conversion

Grapheme-to-phoneme conversion predicts the sounds represented by written text, supporting text-to-speech, pronunciation tools, and phonetic transcription. Systems range from dictionaries and language-specific rules to neural sequence models and large language models, sometimes combining pronunciation with word segmentation or sentence context. Because labeled pronunciation data are scarce for many languages, researchers also use weak labels from dictionaries, generated text, or speech audio, and evaluate systems with phone error rates and language-specific benchmarks. A key challenge is balancing more training data and broader language coverage with reliable handling of dialects, ambiguous spellings, and nonstandard user-generated text.