Voice Recognition in HardwareAs an Input DeviceTao Ben SchardlNicholas HarringtonNicholas Harrington & Tao Ben Schardl () Voice Recognition November 13, 2008 1 / 11A Functional OverviewSmall sections of incoming audio (≈ 20ms ) are analyzed andconverted into vectors that parameterize the incoming audio.Each word is composed of a list of these vectors. Endpoints forpossible words are determined by the overall energy of incoming audio.Words trained into the hardware are compared against incoming audiosignals using a dynamic time warping alogorithm (DTW).Nicholas Harrington & Tao Ben Schardl () Voice Recognition November 13, 2008 2 / 11Vector GenerationNicholas Harrington & Tao Ben Schardl () Voice Recognition November 13, 2008 3 / 11Vector GenerationBorrowed from wikipediaNicholas Harrington & Tao Ben Schardl () Voice Recognition November 13, 2008 4 / 11DTWAlgorithmei(j+1)= di(j+1)+min(eij, ei(j−1), e(i−1)(j−1))dij= d (wi, sj−i)Nicholas Harrington & Tao Ben Schardl () Voice Recognition November 13, 2008 5 / 11DTWNicholas Harrington & Tao Ben Schardl () Voice Recognition November 13, 2008 6 / 11DTW ControlNicholas Harrington & Tao Ben Schardl () Voice Recognition November 13, 2008 7 / 11Mock LayoutNicholas Harrington & Tao Ben Schardl () Voice Recognition November 13, 2008 8 / 11Output ModuleNicholas Harrington & Tao Ben Schardl () Voice Recognition November 13, 2008 9 / 11TimelineNicholas Harrington & Tao Ben Schardl () Voice Recognition November 13, 2008 10 / 11Thank You For Your TimeNicholas Harrington & Tao Ben Schardl () Voice Recognition November 13, 2008 11 /
View Full Document