Atmosphere

What it does

Atmosphere interacts with the Google Gemini Flash API in 2 independent steps to generate immersive and cohesive audio recordings.
Step 1: Timestamps
Initially, Atmosphere is provided with the audio recording in its entirety, and is tasked with locating timestamps within the audio book that correspond to scenes that would be enhanced by ambient audio. In addition to providing the timestamp of these scenes, Gemini also produces a concise description of the context of the scene, and salient keywords from the scene that construct the overall tone of the segment.
Step 2: Mappings
Using the keywords gathered for each scene identified by Gemini, Atmosphere sifts through the 33,000+ sound effects and their associated keywords in the BBC sound effects library, and acquires all sounds with at least one keyword matching that of the scene. With all sound effects a minimum of one matching keyword compiled into a list, Gemini is prompted to pick the sound from the list that best matches the scene description created by Gemini in step 1.
Step 3: Overlay
Once all sounds have been selected for each scene, Atmosphere normalizes, fades, trims then overlays the selected sound effects onto the corresponding audio book segments.