Jonathan Wax, General Manager, EMEA, Nexidia, Ltd., talked with me today about how the difficulties of converting spoken words to searchable text can be solved by using phonemes.
To illustrate his point he did a search on BBC videos for “Saddam Hussein.”
“It doesn’t matter how you spell it,” he said. (Lucky for the spelling-challenged like me.)
Once translated to phonemes, the phoneme string can be searched against audio tracks translated by the same method.
The demo did indeed rapidly retrieve some BBC clips tagged to the point where the reference to “Sad am Hu zane” occurred. In fact, we zipped directly to a BBC clip on Hussain’s recent sentencing.
The technology behind the method emerged from research at Georgia Tech.
Besides obvious applications for indexing and searching audio files, Wax said the technology is also being used in intelligence applications.
ITI, V.P. Content