Decoding

1
Decoding Playing Digital Speech

• COEN339 Course Project
• Spring 2002
• Members Julia Sulisthio, Bill Yi
• Professor Dr. Nam Ling

2
Project Overview

• Digital speech decoder G.729 Annex A
• Decompression technique CS-ACELP
• Performance issues MOS
• Implementation experimental results
• Conclusion questions

3
Digital Speech Decoder G.729A

• A wind of change in the modem industry!
• Overview of G.729 Annex A
• 8 kbps DSVD Digital Simultaneous Voice Data
• ACELP technology A real Revolution
• For wireless and multimedia network applications
• Benefits of using G.729A
• Low complexity MIPS, RAM, ROM
• Low delay 10 5 15 ms total
• Toll quality equal/higher than 32 kbps ADPCM
• Interoperability compatible with G.729

4
Decompression Technique

• CS-ACELP Conjugate-Structure Algebraic-Code-Exci
  ted Linear Prediction
• Parameters LP filter coefficients, adaptive and
  fixed-codebook indices gains

5
Performance Issues - MOS
MOS Codec Mean Opinion Score DRT Diagnostic
Rhyme Test DAM Diagnostic Acceptability Measure
6
Implementation

• Sample program
• Intel Pentium II 350 processor
• Source g729a_encoder.c, g729a_decoder.c, etc.
• Codec complexity Encoder 5, Decoder 3 CPU time
• Procedure
• sox r 8000 w s f1n_org_nb.wav speech_f.raw
• g729a_encoder speech_f.raw bitstream
• g729a_decoder bistream speechout_f.raw
• sox r 8000 w s speechout_f.raw speechout_f.wav

7
Experimental Results

• Windows Media Player Sound Recorder

(a) Original speech at 1.21 sec
(c) Original speech at 12.92 sec
(b) Reconstructed speech
(d) Reconstructed speech
8
Conclusion Questions

• Coding is an important component for voice
  applications
• G.729A using CS-ACELP towards signal uniformity
  and compatibility ? G.729A is the answer
• Room for improvement in existing algorithm
• Save more bits
• Full toll quality
• Time for Questions??