Glossary
Spontaneous vs read speech
Spontaneous speech is unscripted natural talk, complete with fillers, interruptions, and real prosody; read speech is scripted and clean. Podcasts and interviews supply the spontaneous kind, which is scarcer in existing datasets.
The differences are structural: spontaneous talk has fillers, restarts, interruptions, variable pace, and real conversational prosody; read speech is fluent and even. Systems trained mostly on read or scripted speech degrade on real conversation. Classic spontaneous-speech corpora are decades old and small next to modern needs, which keeps the register scarce.
Why it matters
Podcasts, interviews, and meetings hold the register models still find hardest.