Speech Recognition SystemJaime DíazRaiza MuñizSystem OverviewClosed-Set Speaker ID systemTwo active statesSpeaker trainingSpeaker IDProject partitionJaime – DSP (feature extraction & comparison).Raiza – control, memory & video output.Block DiagramExtractorDistanceMemoryControl UnitRegisterDisplayFeature_compareExtractedFeaturesExtractWriteAddressData_outEnableEnableDoneDistanceSyncAdd_UserIDReset (To all blocks)IDAction Reset_Reg2UserLine_CountPixel_CountRGBVGAVGA_OutVoiceResetAdd_UserExtractor BlockProcesses ~ 3.5 sec audioOutputs 16 Spec. Coeff.Issue: number of samplesNeed to process small chunksPipelining to reduce gatesCustomization less portableHammingWindowDFTMel FiltersDCTAC’97LogVoiceSpectral CoefficientsDistance BlockCompares Spec. Coef. (SC)Input vs Stored (Speech)Outputs a distance metricComparison: Dynamic Time WarpingCalc. Euclidean distance bet the SC of input vs stored for each time interval.Dist = Σ smallest dist in each TI row and column of the distance matrix.-XXIIISS-------4-2-3-9S698988215677576S7S8I8I8X7X9X7X4-3-4-Distance Metric Calculation ExampleControl BlockTells all other blocks what to do.Drives the direct user I/O interfaceADD or ID user inputs.Video outputDrives Memory Read/Write cyclesSupplies Distance Block stored SC vectors.Other BlocksMemory – store/read user SC as neededRegister – tell Control requested actionVideo interface – feedback to the userThank
View Full Document