Soundtrack Search Accuracy
To achieve high accuracy in the search of a video's soundtrack, in any language, requires a faithful textual transcription of every word in the narration and dialogue contained in the soundtrack. The most reliable solution is to take the subtitle or closed caption data that usually accompanies professionally produced video and to import that data into a search engine index.
 
Subtitle data is essentially a set a lines of text and the timing points for when that text should be rendered on screen. Our video search engine consists of an SQL index of all the words within the collection of video to be searched, with each word linked to one or more subtitle lines from all the videos in the collection in which that word occurs in the soundtrack.
 
When a user searches for a word or a phrase, our search engine generates a set of results, ordered by relevance, showing which lines of narrative or dialogue were matched, which individual video they occurred in and the relevant timepoints.
 
Matching Scene Images
The results generated by our video search engine also show thumbnail still images of the scenes in the particular videos in which the matched words occur. We ensure that the still images shown actually correspond to the matched scenes as indicated by the search engine.
 
Precision Timing
A major difficulty with this kind of video search is that the search engine index data is kept in a database which is entirely separate from the actual video and its real soundtrack. If the timing information in the index is not perfectly synchronised with the actual video itself, users might click a search result to play a matched scene and be shown the wrong part of the video.
 
We have developed software techniques to overcome this difficulty by automatically synchronising the timing data in the search engine index with the true video soundtrack at multiple points throughout each video we index. In this way we can guarantee an accuracy of within one second for implementations based on original subtitle data.
 
Subtitle Data Handling
We can take subtitle or closed caption data in any recognised format in text form. We can also take subtitle data sets in the form of stacks of bitmaps (as used on DVD) and convert those to a standardised text format. We then convert the subtitle data into a proprietary data format that we have perfected for use in our video search engine.
 
Rapid Implementation
For implementations where good quality subtitle data is available, we can construct an isolated instance of our search engine for a collection of video of any size in a matter of hours. In most cases we also take care of all the video preparation, encoding, hosting and streaming but we can also integrate a separately hosted instance of our video search engine with an existing video delivery site or application. See our section on custom video search applications for more information.
 
 
To talk to us about incorporating our video search into your business, please email us: sales@doovle.com.