Chinese Speech Recognition System Development
30 likes | 148 Vues
Develop a speech recognition system for Chinese languages using HMM initial models. Includes lexicon lookup, model parameter allocation, MLVQ processes, and state transition probability estimation. Save and assign model parameters efficiently.
Chinese Speech Recognition System Development
E N D
Presentation Transcript
語音訊號 412 0 0 29 313 0 30 31 313 1 32 35 313 2 36 40 313 3 41 42 313 4 43 59 313 5 60 63 313 6 64 66 313 7 67 71 412 0 72 79 Feature , (查cut) 表示成所有中文語言單元 經過lexicon字彙庫找出所有的子音(22)、母音(39)、 左至右HMM initial final 32mixture VQ 32mixture VQ 32個µ、v MLVQ 32個µ、v MLVQ u、v、c、det u、v、c、det
begin 411.c m m • Allocate initial & final model parameters • transition prob • mean • variance • weight • determinant 1 m> F_MODEL m> I_MODEL 1 0 0 s s 1 s> F_STATE 1 s> I_STATE 0 0 Open the database list file (fp_list) Read frames from the cls file Read frames from the cls file Load lexicon • VQ process • get mixture means • VQ process • get mixture means • Remove model’s temp file • initialize • model/state’s count • Initialize MLVQ • determine • variance • weight • Initialize MLVQ • determine • variance • weight • Generate new temp files • all speakers and all utterances • read feature and get delta term • SBR • read cut information • all syllables • all states • write state frames to the • cls file of the • corresponding state • model • update model’s • and state’s count • MLVQ process • get mean, variance weight, determinant • MLVQ process • get mean, variance, weight, determinant Save 411 parameters of HMM model Assign the mixture number of initial and final states Delete old cls files end Estimate the state transition prob