Freesound interprets CC licenses for AI training as requiring dataset disclosure for BY sounds and barring commercial use for BY-NC sounds

Creative Commons licenses predate generative AI and do not explicitly address model training. Freesound’s 2024 interpretation maps existing CC terms onto the AI training context: CC0 sounds can be used to train any model without restriction; CC-BY sounds can be used if the training dataset is publicly disclosed (treating disclosure as equivalent to attribution); CC-BY-NC sounds require dataset disclosure and further that the model is not trained commercially nor used for commercial purposes. These interpretations are Freesound’s own reading, not binding legal rulings. Additionally, from February 2026 individual uploaders can flag their preference for AI use via an account setting, but those preferences are explicitly non-legally-binding. Model developers can query these preferences via the Freesound API.

Examples

A research lab wants to train an open-source audio generation model on Freesound sounds. CC0 sounds: usable freely. CC-BY sounds: usable if the training dataset list is published. CC-BY-NC sounds: usable only if dataset is disclosed and the model is non-commercial.

Assessment

A startup trains a commercial audio generation model on Freesound sounds. Under Freesound’s interpretation, which license tier(s) can they use without violating the licenses? What additional step allows use of CC-BY sounds?