System and method for context detection of objectionable speech in video
AssigneeInternational Business Machines Corporation
InventorsIstvan Rechner, Attila Toth, Andras Balogh, Eszter Kotra
Life of this patent
Estimating the term…
Abstract
Embodiments provide a system and method for filtering speech in a video. Speech in video may contain objectionable words that need to be filtered. To ascertain whether a word or phrase is objectionable, the contextual information from surrounding words and the contextual information from detected objects and scenes in the video are used. Unwanted words may be filtered or collected and presented to the user.
Claims16
A computer implemented method in a data processing system comprising a processor and a memory comprising instructions, which are executed by the processor to cause the processor to implement speech filtration of a video, the method comprising:
- converting speech in a scene of the video to text, wherein the text comprises a plurality of triplets, each triplet includes a word, a start time of the word in the scene, and an end time of the word in the scene;
- determining a language of the text based on a frequency distribution of n-grams of the text, wherein n is any integer;
- selecting a dictionary of objectionable words based on the language of the text;
- comparing each word in the text to the objectionable words in the selected dictionary;
- filtering a first list of candidate words from the text, wherein each filtered candidate word has at least one non-objectionable meaning;
- for each candidate word having at least one non-objectionable meaning:
- identifying surrounding words in the scene;
- calculating a first context for the candidate word using semantic analysis of the surrounding words;
- receiving a list of objects in the scene;
- calculating a second context for the candidate word using semantic analysis of the list of objects;
- and
- replacing the candidate word with a non-objectionable alternative word if the candidate word is objectionable based on the first and second contexts.
A computer program product for speech filtration of a video, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
- convert speech in a scene of the video to text, wherein the text comprises a plurality of triplets, each triplet includes a word, a start time of the word in the scene, and an end time of the word in the scene;
- determine a language of the text based on a frequency distribution of n-grams of the text, wherein n is any integer;
- select a dictionary of objectionable words based on the language of the text;
- compare each word in the text to the objectionable words in the selected dictionary;
- filter a first list of candidate words from the text, wherein each filtered candidate word has at least one non-objectionable meaning;
- for each candidate word having at least one non-objectionable meaning:
- identify surrounding words in the scene;
- calculate a first context for the candidate word using semantic analysis of the surrounding words;
- receive a list of objects in the scene;
- calculate a second context for the candidate word using semantic analysis of the list of objects;
- and
- replace the candidate word with a non-objectionable alternative word if the candidate word is objectionable based on the first and second contexts.
A system for providing speech filtration of a video, comprising a processor configured to:
- convert speech in a scene of the video to text, wherein the text comprises a plurality of triplets, each triplet includes a word, a start time of the word in the scene, and an end time of the word in the scene;
- determine a language of the text based on a frequency distribution of n-grams of the text, wherein n is any integer;
- select a dictionary of objectionable words based on the language of the text;
- compare each word in the text to the objectionable words in the selected dictionary;
- filter a first list of candidate words from the text, wherein each filtered candidate word has at least one non-objectionable meaning;
- for each candidate word having at least one non-objectionable meaning:
- identify surrounding words in the scene;
- calculate a first context for the candidate word using semantic analysis of the surrounding words;
- receive a list of objects in the scene;
- calculate a second context for the candidate word using semantic analysis of the list of objects;
- replacing the candidate word with a non-objectionable alternative word if the candidate word is objectionable based on the first and second contexts.
Dependencies are inferred from the claim text.