Audio scene generation

info

Audio scene generation is currently in beta.

An audio scene is a declarative description of a listening scenario. It places one or more source recordings in a modeled acoustic environment, assigns each recording to an impulse response (IR), and defines how the result is captured by a listener device.

The Treble SDK represents this recipe with an AudioScene. The scene stores the selected source tracks, source-to-IR mappings, source grouping, listener configuration, and duration. Rendering happens later, when audio output is requested from the scene.

Scene model

Audio scene generation combines three kinds of input:

• room acoustics from an IRCollection
• source recordings such as speech, background noise, natural sounds, or music
• listener configuration such as device, orientation, device noise, and output filters

The result is a structured scene definition that can be inspected, serialized, rendered, and queried for metadata.

Workflow types

The scene tools support two complementary workflows:

• Manual scene construction, where each AudioScene is defined directly by assigning IRs, track content, and listener configuration.
• Bulk scene generation, where reusable SceneRules are passed to SceneGenerator to produce a randomized SceneCollection.

The manual workflow supports precise control over one scene. The bulk workflow supports large sets of varied scenes for dataset generation, benchmarking, or algorithm evaluation.

Rendering lifecycle

An AudioScene doesn't contain rendered waveforms when it is first created. It contains enough information to retrieve the required IRs and source recordings later, apply leveling and filtering, convolve tracks through the selected IRs, and mix the final output.

For task-focused examples, see Manual scene generation, Bulk scene generation, and Rendering a scene collection.