Glossary
Phoneme coverage
How completely a speech dataset represents the distinct sound units of a language — important for pronunciation and TTS quality.
English has about 40 distinct phonemes, depending on dialect and analysis, and corpora are checked for how often each appears — and in how many neighbouring-sound contexts, since pronunciation shifts with context. Scripted collections balance coverage deliberately; conversational archives reach it through volume and variety. Gaps surface later as words a synthetic voice mispronounces.
Why it matters
A dataset spec that documents phoneme balance answers one of the first questions TTS teams ask.