Data Compression


Published on

This presentation describing speech synthesis methods.

Published in: Education, Technology, Business
1 Like
  • Be the first to comment

No Downloads
Total Views
On Slideshare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide
  • Draw Pictures on a board here.
  • Data Compression

    1. 1. By Ashwin Pannikar and Sanket Yavalkar
    2. 2. <ul><li>Why Fractal Image Compression </li></ul><ul><li>How does it work? </li></ul><ul><li>Examples </li></ul><ul><li>Possible Improvements </li></ul><ul><li>Channel Vocoders </li></ul><ul><li>LPC-10 </li></ul>
    3. 3. <ul><li>Different type of compression scheme worth exploring </li></ul><ul><li>Takes advantage of similarities within an image </li></ul><ul><li>Advanced detail interpolation </li></ul><ul><li>High theoretical compression rates </li></ul><ul><li>Fast decompression times </li></ul>
    4. 4. <ul><li>Take a starting image and divide it into small, non-overlapping, square blocks, typically called “parent blocks”. </li></ul><ul><li>Divide each parent block into 4 each blocks, or “child blocks.” </li></ul><ul><li>Compare each child block against a subset of all possible overlapping blocks of parent block size. </li></ul><ul><ul><li>Need to reduce the size of the parent to allow the comparison to work. </li></ul></ul><ul><li>Determine which larger block has the lowest difference, according to some measure, between it and the child block. </li></ul><ul><li>Calculate a grayscale transform to match intensity levels between large block and child block precisely. Typically an affine transform is used (w*x = a*x + b) to match grayscale levels. </li></ul>
    5. 5. <ul><li>Upper left corner child block, very similar to upper right parent block. </li></ul><ul><li>Compute affine transform. </li></ul><ul><li>Store location of parent block (or transform block), affine transform components, and related child block into a file. </li></ul><ul><li>Repeat for each child block. </li></ul><ul><li>Lots of comparisons can calculations. </li></ul><ul><ul><li>256x256 original image </li></ul></ul><ul><ul><li>16x16 sized parent blocks </li></ul></ul><ul><ul><li>241*241 = 58,081 block comparisons </li></ul></ul>
    6. 6. <ul><li>Read in child block and tranform block position, transform, and size information. </li></ul><ul><li>Use any blank starting image of same size as original image </li></ul><ul><li>For each child block apply stored transforms against specified transform block </li></ul><ul><li>Overwrite child block pixel values with transform block pixel values </li></ul><ul><li>Repeat until acceptable image quality is reached. </li></ul>
    7. 7. <ul><li>Original Image </li></ul><ul><li>Starting Image for Decoding </li></ul>
    8. 8. <ul><li>First Iteration </li></ul><ul><li>Second Iteration </li></ul>
    9. 9. <ul><li>Fifth Iteration </li></ul><ul><li>Tenth Iteration </li></ul>
    10. 10. <ul><li>Greatest weakness is time for encoding </li></ul><ul><ul><li>Possible speed ups </li></ul></ul><ul><ul><ul><li>Order transform blocks into domains based off of average intensity and variance </li></ul></ul></ul><ul><ul><ul><li>Only search through blocks with similar structures </li></ul></ul></ul><ul><ul><ul><li>Do not search all possible blocks </li></ul></ul></ul><ul><ul><ul><li>Reduce number of child blocks </li></ul></ul></ul><ul><li>Quality and Compression Improvements through </li></ul><ul><ul><li>Quadtrees or HV Trees </li></ul></ul><ul><ul><li>Rotations of Transform Blocks during comparison </li></ul></ul><ul><ul><li>Improved grayscale transforms </li></ul></ul>
    11. 12. <ul><li>The channel vocoder employs a bank of bandpass filters, </li></ul><ul><ul><li>Each having a bandwidth between 100 HZ and 300 HZ. </li></ul></ul><ul><ul><li>Typically, 16-20 linear phase FIR filter are used. </li></ul></ul><ul><li>The output of each filter is rectified and lowpass filtered. </li></ul><ul><ul><li>The bandwidth of the lowpass filter is selected to match the time variations in the characteristics of the vocal tract. </li></ul></ul><ul><li>For measurement of the spectral magnitudes, a voicing detector and a pitch estimator are included in the speech analysis. </li></ul>
    12. 13. Bandpass Filter A/D Converter Lowpass Filter A/D Converter Lowpass Filter Rectifier Rectifier Bandpass Filter Voicing detector Pitch detector Encoder S(n) To Channel
    13. 14. <ul><li>At the receiver the signal samples are passed through D/A converters. </li></ul><ul><li>The outputs of the D/As are multiplied by the voiced or unvoiced signal sources. </li></ul><ul><li>The resulting signal are passed through bandpass filters. </li></ul><ul><li>The outputs of the bandpass filters are summed to form the synthesized speech signal. </li></ul>
    14. 15. D/A Converter Decoder D/A Converter Voicing Information Pitch period Pulse generator Random Noise generator Bandpass Filter Bandpass Filter Switch ∑ From Channel Output speech
    15. 16. <ul><li>The objective of LP analysis is to estimate parameters of an all-pole model of the vocal tract. </li></ul><ul><li>Several methods have been devised for generating the excitation sequence for speech synthesizes. </li></ul><ul><li>LPC-type of speech analysis and synthesis are differ primarily in the type of excitation signal that is generated for speech synthesis. </li></ul>
    16. 17. <ul><li>This methods is called LPC-10 because of 10 coefficient are typically employed. </li></ul><ul><li>LPC-10 partitions the speech into the 180 sample frame. </li></ul><ul><li>Pitch and voicing decision are determined by using the AMDF and zero crossing measures. </li></ul>