Sung audio for Emotion & prosody

Singing is emotion made deliberate: vibrato depth, dynamics, breathiness, timing rubato — the acoustic levers of affect, performed on purpose and to an extreme spoken corpora rarely reach. That makes sung audio a dense training signal for models that read or generate expressive audio. It stretches the feature space: a model that has only heard conversational arousal has never seen the sustained high-energy phonation or controlled pianissimo that singing supplies.

Labelling differs from speech emotion work: useful units are phrases and sections rather than turns, and annotation can lean on musical structure — dynamics markings, key, tempo — as weak labels alongside human ratings of expressive intent. The standing caveat: performed emotion is stylized. Corpora should label expressive technique and intended affect separately, so models do not learn that vibrato equals sadness.

Industry-typical ranges — a brief can and should deviate where the task demands it.

The shape of a workable request — swap in your own numbers and conditions:

• Modality: solo vocal phrases performed at controlled expressive intents.
• Volume: 20 hours; same material sung at contrasting dynamics and affect.
• Labels: intent tags, technique tags, 3-rater valence/arousal per phrase.
• Rights: performer consent naming AI training; original material only.

Music rights are famously tangled. fiund only lists sung audio where the performer and rights are documented — a small but genuinely clean supply. We source to a brief and clear the rights before anything moves, so what you receive is both useful and defensible in diligence.

Signed licence, explicit training rights, separate voice/likeness consent, nothing scraped. See the rights & provenance guides.

← All Sung audio data