Applications and Derivation of Linear Predictive Coding

1,068 views

Published on

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
1,068
On SlideShare
0
From Embeds
0
Number of Embeds
2
Actions
Shares
0
Downloads
50
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Applications and Derivation of Linear Predictive Coding

  1. 1. linear predictive coding (LPC) an application driven approach adapted from guest lecture for mobile application development for sensing and control, EE596 Friday, August 30, 13
  2. 2. non-parametric parametric Use Data or Transform Fit Data to a Model Data Derivative Friday, August 30, 13
  3. 3. non-parametric parametric Use Data or Transform Fit Data to a Model Data Derivative Friday, August 30, 13
  4. 4. non-parametric parametric Use Data or Transform Fit Data to a Model Data Derivative derivative[n] = y[n]-y[n-1] Friday, August 30, 13
  5. 5. non-parametric parametric Use Data or Transform Fit Data to a Model Data Derivative derivative[n] = y[n]-y[n-1] Friday, August 30, 13
  6. 6. non-parametric parametric Use Data or Transform Fit Data to a Model Data Derivative derivative[n] = y[n]-y[n-1] Friday, August 30, 13
  7. 7. non-parametric parametric Use Data or Transform Fit Data to a Model Data Derivative cos(x) derivative[n] = y[n]-y[n-1] Friday, August 30, 13
  8. 8. non-parametric parametric Use Data or Transform Fit Data to a Model Data Derivative cos(x) derivative[n] = y[n]-y[n-1] -sin(x) Friday, August 30, 13
  9. 9. the tradeoff of parametric modeling Friday, August 30, 13
  10. 10. the tradeoff of parametric modeling - need to fit a model to the data Friday, August 30, 13
  11. 11. the tradeoff of parametric modeling - need to fit a model to the data + (might be) easier to manipulate model Friday, August 30, 13
  12. 12. non-parametric parametric Use Data or Transform Fit Data to a Model two signals = 1500 Hz and 5500 Hz two signals = 1500 Hz and 5500 Hz Magnitude freq (kHz) freq (kHz) Magnitude Friday, August 30, 13
  13. 13. non-parametric parametric Use Data or Transform Fit Data to a Model two signals = 1500 Hz and 5500 Hz FFT, array two signals = 1500 Hz and 5500 Hz Magnitude freq (kHz) freq (kHz) Magnitude Friday, August 30, 13
  14. 14. non-parametric parametric Use Data or Transform Fit Data to a Model two signals = 1500 Hz and 5500 Hz FFT, array two signals = 1500 Hz and 5500 Hz Magnitude freq (kHz) LPC polynomial freq (kHz) Magnitude Friday, August 30, 13
  15. 15. what model should we fit to? Friday, August 30, 13
  16. 16. what model should we fit to? a filter with feedback Friday, August 30, 13
  17. 17. what model should we fit to? a filter with feedback Friday, August 30, 13
  18. 18. what model should we fit to? a filter with feedback Friday, August 30, 13
  19. 19. feedback filters are system models Friday, August 30, 13
  20. 20. feedback filters are system models Friday, August 30, 13
  21. 21. feedback filters are system models Friday, August 30, 13
  22. 22. feedback filtering a Friday, August 30, 13
  23. 23. feedback filtering want to estimate a a Friday, August 30, 13
  24. 24. feedback filtering what can we represent with this equation? Friday, August 30, 13
  25. 25. ak k k k feedback filtering what can we represent with this equation? 3 2 1 0 -1 -2 -3 1 3 5 7 9 11 13 3 2 1 0 -1 -2 -3 1 3 5 7 9 11 13 3 2 1 0 -1 -2 -3 1 3 5 7 9 11 13 piano marimba violin Friday, August 30, 13
  26. 26. feedback filter equation in frequency … Friday, August 30, 13
  27. 27. feedback filter equation in frequency … Y (z) = E(z) 1 Pp k=1 akz k z = ej! Friday, August 30, 13
  28. 28. is this a good model for frequency analysis? Y (z) = 1 1 Pp k=1 akz k E(z) Y (z) = 1 Qp k=1(1 rkz 1) E(z) Friday, August 30, 13
  29. 29. is this a good model for frequency analysis? resonant frequency = complex angle of root resonance bandwidth = related to magnitude of root Y (z) = 1 1 Pp k=1 akz k E(z) Y (z) = 1 Qp k=1(1 rkz 1) E(z) Friday, August 30, 13
  30. 30. examples Y (z) = 1 Qp k=1(1 rkz 1) E(z) Friday, August 30, 13
  31. 31. another interpretation, vocal tract sourcefilter Y (z) = 1 1 Pp k=1 akz k E(z) Friday, August 30, 13
  32. 32. another interpretation, vocal tract sourcefilter Y (z) = 1 1 Pp k=1 akz k E(z) Friday, August 30, 13
  33. 33. another interpretation, vocal tract sourcefilter Y (z) = 1 1 Pp k=1 akz k E(z) Friday, August 30, 13
  34. 34. another interpretation, prediction … Friday, August 30, 13
  35. 35. another interpretation, prediction … Friday, August 30, 13
  36. 36. 17 Friday, August 30, 13
  37. 37. 18 Friday, August 30, 13
  38. 38. 18 Friday, August 30, 13
  39. 39. 18 Friday, August 30, 13
  40. 40. summary of interpretations Friday, August 30, 13
  41. 41. summary of interpretations Spectral Estimation == Auto Regressive Friday, August 30, 13
  42. 42. summary of interpretations Spectral Estimation == Auto Regressive Forecasting == Linear Prediction Friday, August 30, 13
  43. 43. summary of interpretations Spectral Estimation == Auto Regressive Forecasting == Linear Prediction Vocal Tract Model == Source/Filter Friday, August 30, 13
  44. 44. common applications Friday, August 30, 13
  45. 45. common applications Speech Vocoders Friday, August 30, 13
  46. 46. common applications Speech Vocoders Spectral Analysis Friday, August 30, 13
  47. 47. common applications Speech Vocoders Spectral Analysis Pitch Estimation Friday, August 30, 13
  48. 48. common applications Speech Vocoders Spectral Analysis Pitch Estimation Voice Changers Friday, August 30, 13
  49. 49. common applications Speech Vocoders Spectral Analysis Pitch Estimation Voice Changers Friday, August 30, 13
  50. 50. common applications Speech Vocoders Spectral Analysis Pitch Estimation Voice Changers Analysis/Synthesis of Instrument Sounds Friday, August 30, 13
  51. 51. common applications Speech Vocoders Spectral Analysis Pitch Estimation Voice Changers Analysis/Synthesis of Instrument Sounds Voice Box Friday, August 30, 13
  52. 52. common applications Speech Vocoders Spectral Analysis Pitch Estimation Voice Changers Analysis/Synthesis of Instrument Sounds Voice Box Compression (i.e., mpeg4, CELP) Friday, August 30, 13
  53. 53. common applications Speech Vocoders Spectral Analysis Pitch Estimation Voice Changers Analysis/Synthesis of Instrument Sounds Voice Box Compression (i.e., mpeg4, CELP) My research– medical sensing from a microphone Friday, August 30, 13
  54. 54. questions? Topics Related to LPC and Further Reading: LPC10, Ultra Low Bit Rate Voice Coding Code Excited Linear Prediction Levinson-Durbin Recursion Burg’s Method LP Cepstral Coefficients The Talking Orchestra SpiroSmart, the mobile phone spirometer eclarson.com eclarson@uw.edu @ec_larson electrical engineering computer science Friday, August 30, 13

×