A tool for real time Keyword Spotting in phone conversations.
The keyword spotting technology (KWS) enables the detection of word occurrences in audio or video streams. The detection consists in computing a score for word utterances. When the score exceeds a threshold, the time window in which the word is detected is stored. The result of KWS is a list of word occurrences.
KWS requires less resources than Automatic Speech Recognition (ASR) and can be easily integrated in embedded systems. Furthermore, KWS works better in low resolution / noisy audio files.
The area of application for KWS is given by developing more complex functionalities starting from the base functionality. The KWS can follow the utterance of a specific word or a specific set of words in any or given order by time line aligning the results obtained by searching a set of words in the same audio stream. On the other hand, each word or combination of words can trigger an action like record the occurrence to be later use or start a specific task.
Automatic monitoring of media streams such as TV shows or radio broadcasts for certain keywords allows very powerful statistics to be gathered. Air-time for specific people or products can be quantified, certain word combinations can be identified. Furthermore, detecting words and their time index allows powerful text analytics to be applied, such as data mining or sentiment analysis on media files.
Automated answering robots benefit from keyword detection as important topics are spotted immediately irrespective of the way the sentence is formulated, which might be a problem for speech recognition. Time savings are made by transferring the caller to the appropriate department or operator.
HID can benefit from keyword detection technology in uncommon contexts when the ability of the operator interaction is limited for various reasons, such as task magnitude or human disabilities.
For certain speech interaction scenarios where short definite answers are required (such as in confirmations for example), keyword detection performs better than recognizing questions or answers that contain interjections and other audio place-holders.
Automatic detection of certain keywords significantly boosts media classification accuracy. As these files are tagged with detected keywords, they can subsequently be readily indexed and retrieved by any search system.
Systems can use keyword detection as an authentication method that actually increases security. By detecting a set of ordered keywords and not an entire sentence, such a system allows its user to speak a different sentence each time, containing only a target set of required keywords.
Keyword detection on mobile devices enabling users to have a fully hands-free experience by listening continuously for specific keywords to initiate voice input.
Keyword detection enables very fast search of interesting words in audio streams.
Detecting target words in movies or other audio-visual media is done without needing to perform Automatic Speech Recognition that might not perform adequately in noisy sound scenarios.