Speech Recognition - A Fresh Approach
Summit Middle School
Floor Location : J 024 N
In in this project it is discussed the current state of the art of speech recognition algorithms and why is the progress in this field so slow, for example Apple's Siri or Google's speech recognition does not work without internet connection. This is because both of them require transferring what you said to their data center. This project offers a new promising approach to speaker independent sound recognition that can lead to 100% accuracy of continuous speech recognition and doesn't require a lot of computer resources. Currently along this way this project achieved very accurate speaker independent recognition of few sounds out of all possible signals. I developed a test bed software that does this recognition real time consuming very little computer resources, and this project is going to use this software in our future work on this topic. These results convince us that it is possible to extend this approach to all human generatable sounds. So after our algorithm can hear/recognize what people say, we can use search algorithms that at this time are well developed or more advanced associative memory algorithms to recognize words out of our recognized sounds. So this way this project can achieve 100% accuracy of speaker independent continuous speech recognition.