International Business Machines Corporation
Disambiguation of audio content using visual context

Abstract:

Provided is a method for disambiguating an audio component extracted from audiovisual content. Audiovisual content is identified. The audiovisual content includes an audio component and a video component. An ambiguous expression is detected in the audio component. An object referenced by the ambiguous expression is identified in the video component. A verbal description of the object is generated. The verbal description is injected into the audio component to generate a modified audio component.