Techniques to provide content based on detected ambient media are disclosed. In various embodiments, a set of audio features extracted from audio data associated with an audio environment with which a client device is associated are received from the client device. A stored media signature data generated based on media content data from a source other than the client device is used to detect, based at least in part on the set of audio features, a media channel and program that is being played in the audio environment. A secondary content that is selected based at least in part on a semantic content data associated with the detected media channel and program is provided.