Backward Compatible Wideband Voice over Narrowband LowResolution Media

1
Backward Compatible Wideband Voice over
Narrowband Low-Resolution Media
Heping Ding, Acoustics and Signal Processing
Group, IMS Feb. 14, 2005

• Problem and prior art
• Our goal
• Implementation schemes CR, MP, BM
• Application Audio Bandwidth Extension

2
Problem and Prior Art

• Standard telephone network
• is meant to transmit toll-quality (barely
  acceptable) voice only
• has a narrow band (NB, 300 - 3400 Hz), used up by
  voice
• has little room in bandwidth/capacity extension
  in order to incorporate voice quality improvement
  and/or additional services

3
Problem and Prior Art

• Existing ways of extending the capacity
• Time- or frequency-division, e.g., call waiting,
  CLID
• Simple and widely used
• Causes voice interruption or degradation
• Limited capacity
• Vocoding, e.g., voice audio coding/decoding
  schemes
• Complicated
• Transmits digital signals, not compatible with
  POTS
• SVD, e.g., dial-up modems
• Needs modem hardware - complicated and costly
• Signals transmitted not compatible with POTS

POTS plain ordinary telephone set
4
Our Goal

• Wanted a capacity extending scheme that
• is simple, without the need for special hardware,
  e.g., a modem
• is compatible with existing end-user equipment,
  i.e., a POTS should still be able to access the
  basic voice service
• allows terminals equipped with the decoding
  mechanism to access the added payload

POTS plain ordinary telephone set
5
Implementation Scheme 1 CR

• Component Replacement (CR)

Original NB voice spectrum with perceptual mask
shown
f (Hz)
NB
300
3400
6
Implementation Scheme 2 MP

• Magnitude Perturbation (MP)

Original NB signal spectrum
f (Hz)
NB
300
3400
7
Application Audio Bandwidth Extension

• Bandwidth extension beyond conventional NB (300 -
  3400 Hz) can provide significant improvements in
  intelligibility and voice quality
• Upper band (UB)
• Lower band (LB)
• However, we cant extend the physical bandwidth
  of the existing telephone network beyond NB
• No sweeteners

8
Application Audio Bandwidth Extension
Example 1 Using magnitude perturbation
f
Transmitter
NB
Determine perceptual mask
Perturb NB magnitudes to embed
Network
Band split
LUB
Retain only those above mask
9
Implementation Scheme 3 BM
Bit Manipulation (BM)

• Suitable for digital systems, e.g., linear or
  G.711 formatted data
• a m-law data sample
• Certain M0 can be manipulated without seriously
  harming audio
• Challenges
• Minimize information amount (No. of bits) to
  represent UB
• Maximize capacity of hidden channel

10
Application Audio Bandwidth Extension
Example 2 Using coding-assisted bit manipulation
- transmitter

• Need only part of voice encoder, for spectral
  shape, e.g., LSP coefficients
• Less than 0.3 bit per sample is needed to
  transmit UB information
• Embedding is done in a least audible way by
  modifying an LSB every 3 or 4 samples

11
Application Audio Bandwidth Extension
Example 2 Using coding-assisted bit manipulation
- receiver

• Fast converging adaptive LPC analysis on received
  NB signal to derive excitation
• Need only part of voice decoder, for implementing
  all-pole model

12
Application Audio Bandwidth Extension
Example 2 Using coding-assisted bit manipulation
- demo
1. Original wide-band (16-bit linear, Fs 16
kHz) 2. Narrow-band, up to 3.4 kHz (8-bit
m-law, Fs 8 kHz) 3. Narrow-band
with upper-band information embedded
(8-bit m-law, Fs 8 kHz) 4.
Restored wide-band (16-bit linear, Fs 16 kHz)
13
Application Audio Bandwidth Extension
Possible ways of implementation
NB telephone network
Encoder Decoder