Prosody

Prosody refers to the rhythm, stress, intonation, pitch contour, and pacing of spoken language — the musical layer of speech that sits above individual phonemes (sounds) and conveys meaning, emotion, emphasis, and grammatical structure. Prosody is what makes the same sentence sound like a question ("You're going?") versus a statement ("You're going.") versus excited ("You're going!"). It includes stress patterns (which syllables are emphasised), rhythmic structure (how words group into phrases), and intonation contour (pitch rising and falling across a phrase). In text-to-speech systems, prosody modelling is typically the hardest part — phoneme pronunciation is well-understood, but knowing where to pause, where to rise in pitch, and where to add emphasis requires the model to understand sentence-level meaning. Neural TTS systems learn prosody implicitly from training data; SSML lets developers override prosodic decisions manually with tags like `<prosody rate="slow">` or `<break>` for pauses. Good prosody is what separates "natural-sounding" TTS from "robotic-sounding" TTS.

How it works

Linguistically, prosody has three main components: pitch (fundamental frequency contour of the voice), duration (how long each syllable is held), and loudness (relative intensity). Languages differ in which of these carries the most meaning — English uses pitch heavily for question-vs-statement distinction, Mandarin uses pitch for lexical tone (same syllable means different things at different pitches), Japanese uses duration distinctions critically. Indian languages broadly use all three, with additional complexity in how Devanagari conjuncts and sandhi affect rhythmic structure. Prosody also changes by register and context — news reading uses steady rhythm and falling intonation, excited YouTube commentary uses rising pitch and compressed phrasing, meditation uses deliberate slowing and strategic pauses. TTS platforms that ship tone presets (VoisLabs, ElevenLabs, Murf) are essentially shipping prosodic configurations optimised for content categories.

Examples

Related terms

Learn more

Frequently Asked Questions