Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Speaker Verificationusing Oguma Histogram  S117036   Ami Inoamta  Supervisor:M.Sugiyama
Outline●   Background●   Speaker verification algorithm●   Oguma histogram●   Speech analysis condition●   Learn histogram...
Background●   Diffusion of smartphone.●   Security is more important.    →Speaker verification system●   VQ : most popular...
Speaker verification algorithm         Figure1: Speaker verification algorithm.
Oguma histogram calculation                                     ●   Do not use VQ.                                     ●  ...
Speech analysis condition        Database                   TIMIT     Dialect region             New England       Head co...
Learn histogram,Verification histogram               Figure3:Histogram.
Result of speaker verification           Figure4:Result(frame shift 8ms).           Figure5:Result(frame shift 16ms).
Upcoming SlideShare
Loading in …5
×

S1170136 week9

128 views

Published on

  • Be the first to comment

  • Be the first to like this

S1170136 week9

  1. 1. Speaker Verificationusing Oguma Histogram S117036 Ami Inoamta Supervisor:M.Sugiyama
  2. 2. Outline● Background● Speaker verification algorithm● Oguma histogram● Speech analysis condition● Learn histogram , Verification histogram● Result of speaker verification
  3. 3. Background● Diffusion of smartphone.● Security is more important. →Speaker verification system● VQ : most popular. But calculation amount is large.● Oguma histogram : calculation amount is small. →smartphone application.● First, test this performance in PC terminal.
  4. 4. Speaker verification algorithm Figure1: Speaker verification algorithm.
  5. 5. Oguma histogram calculation ● Do not use VQ. ● Directly make histogram from feature vectors. 1. Set out threshold in each dimension of feature vectors. 2. Compare →Space division 3.Set out Region ID .Figure2:Concept of Oguma histogram
  6. 6. Speech analysis condition Database TIMIT Dialect region New England Head count 10 person Learning , verification 5 sentence , 5 sentence Recording format wav Sampling rate 16 kHzDimension number of MFCC 16 Filter bank channel 24 Window size 16 ms Frame shift 8 ms, 16 ms Dimension number of 16, 32, 64, histogram 128,256,512,1024,2048, 4096,8192
  7. 7. Learn histogram,Verification histogram Figure3:Histogram.
  8. 8. Result of speaker verification Figure4:Result(frame shift 8ms). Figure5:Result(frame shift 16ms).

×