Catalogue of Artificial Intelligence Techniques
Automatic Speech Recognition
Categories: Natural Language
Author(s): Andrew Todd
Automatic Speech Recognition is a process which involves the conversion of speech input, taken usually through microphones, and converting it into a text format. This process has been developed extensively through the 80's and 90's. It typically involves use of Hidden Markov Models, used to filter out noise. The Hidden Markov Model works to assign the most probable text phrase that could be associated with the input sound signal. The algorithm's success is measured based on word accuracy, and the amount of time it takes to produce a result.
Of course, the hidden markov model is not the only approach. Neural Network-based approaches have been tried, but they do not work as well when larger vocabularies are used. Dynamic Time Warping based speach recognition has also been used, since Dynamic Time Warping can detect patterns in speech no matter the rate of at which words are spoken.
- Wikipedia.org, Automatic Speech Recognition Wikipedia.org.