Chap 3' AnalysisbySynthesis - PowerPoint PPT Presentation

1 / 10
About This Presentation
Title:

Chap 3' AnalysisbySynthesis

Description:

Vocal tract model All-pole model (LPC model) good for speech except nasal, ... Spectral Tilt compression add a ploe in AR model. 10. LPT. LPC- Parameters ... – PowerPoint PPT presentation

Number of Views:21
Avg rating:3.0/5.0
Slides: 11
Provided by: speechCm
Category:

less

Transcript and Presenter's Notes

Title: Chap 3' AnalysisbySynthesis


1
Chap 3. Analysis-by-Synthesis
  • Speech generated model
  • LPC model of speech

2
  • Vocal tract model All-pole model (LPC model)
    good for speech except nasal,
  • Glottal model impulse train/noise,

VQ the parameters, transmit them
3
  • Performance of LPC analysis
  • short time prediction short-time stationary
    assumption

4
  • Periodic assumption of speech signal
  • Long term prediction (LTP)
  • Long term prediction Gain
  • Pitch estimation

5
  • Perceptual error weighting filter
  • larger signal power will allow larger noise
    power

6
  • Block diagram
  • A(z) updated frame 10-30ms (80-240 samples, 8KHz
    sampling)
  • Excitation sequence length 5-7.5ms (40-60
    samples)

7
  • Using excitation codebook (save some
    representative excitation waveform)

8
  • Codec Consideration
  • (1) using reflection coefficients in stead of
    LPC coefficients, because rc(k)?1
  • (2) quantization bit number for each coefficient
    (bit-allocation)
  • Excitation (residual) information
  • (1) coding residual signal (using AQ,) RPE
    (Regular pulse excitation)
  • (2) Select the most important (large) excitation
    only MPE (Multi-pulse Excitation)
  • (3) using VQ (Vector Quantizer) to encode
    excitation signal CELP (Code Excitated)

9
  • G.728 Codec
  • Block diagram
  • Spectral Tilt compression add a ploe in AR model

10
  • LPT
  • LPC-
  • Parameters-
  • Bit-rate (bit-allocation is not shown where) 16
    kbps.
Write a Comment
User Comments (0)
About PowerShow.com