What’s new in Wideband

Audio?

Wideband Audio

• VoIP is indeed a disruptive technology, but has it changed the life of the average consumer?– Cost?– Quality?– Features?

• Wideband Audio codecs and improved handling of music could soon change this dynamic

• Let’s discuss– Technology behind the codecs– Real-world implementations

Telecom Audio Spectrum

• Human voice: 80 Hz to 14,00Hz• Narrowband: 8 kHz sampling (300-3400 Hz bandwidth)

– Used in PSTN, mostly intelligible

• Wideband: 16 kHz sampling (50-7000 Hz bandwidth)– Used in VoIP

Wideband Audio?

• Captures significantly more speech information– Significant improvement in speech quality over

traditional PSTN – Improved naturalness & presence below 200Hz– Increased intelligibility above 3,400Hz

• Improves user experience and satisfaction– New applications – voice recognition– Customer retention– Fewer misunderstandings

Wideband Enablers

• Telecom was about minimizing transport cost– Now about differentiation and enhancing the user experience

• Access bandwidth was limited– Broadband access now a reality: high bandwidth delivered at low

cost• 1 - 10 Mbits/s• Cost of WB is similar to NB @ 64kbps

• Endpoints and Network were not wideband capable Now:– VoIP, Wideband DECT, Skype, Microsoft OCS– Wireless deployments: wideband, music codecs– Private / corporate networks, Tandem Free Operation (TFO),

Wideband extension, Wideband SLICS

The Technology

Lossy Codec Classes

• Speech communication codecs (G.72X, AMR et.al)– Designed for “real-time” speech, music handled poorly– Low sampling rate (8-16KHz), low fidelity– Low-medium delay (10-30 ms)– Mostly time-domain (CELP is the most popular)

• Music codecs (MP3, AAC, Vorbis)– Can encode any signal (not optimal for speech) – designed for

entertainment– Up to 48 kHz sampling rate (full bandwidth), high fidelity (“CD-

quality”– High delay (>100 ms)– Mostly frequency domain (MDCT-based)

Speech Codec Spectrum

3.5KhzNarrowband

7KhzWideband

More than 15KhzFull Band (20Khz)

G.729, G.723.1G.711, iSAC

G.722.2 (AMR-WB), SVOPC

AAC-LD

Bandwidth Example CodecApplicationsDeployed

Presence(Video Conf)

BB VoIP & Audio Chat

PSTN &VoIP

14KhzSuper Wideband

VoIP, Audio Conf

G.722.1C (Siren14), SILK

ITU and 3GPP codec roadmap

G.7261984 G.728

1992GSM-FR

1987

GSM-HR

1994

GSM-EFR1995

G.7291995

AMR-NB1999

G.722.2AMR-WB

2002G.7221988

G.722.11999

G.729.12007

narr

owba

nd

EV-VBR2008

wid

eban

d

Years

Supe

r -w

ideb

and

3GPP&

ITUITU 3GPP

Legend:

Embedded Speech Codecs

• ITU-Super WB – Provides extended bandwidth and stereo capabilities– 16 KHz audible bandwidth– Stereo extension– Generic extension applicable to wideband codecs e.g.. ITU

G.729.1 & EV-VBR• 3GPP-EPS (evolved packet system) (aka LTE)

– ITU EV-VBR is well positioned to meet future EPS requirements – Interoperable with 3GPP AMR-WB.

• Open Codecs– Speex (4 to 42Kbps)

• Royalty free but limited to non patented techniques (ACELP for example)

Music Codecs

• MPEG-1 Layer III (aka MP3)– Built on top of Layers I and II– First-generation, very inefficient

• AAC– Second generation, much better than MP3– Flexible, kitchen-sink type of approach– Tons of tools and partially incompatible profiles– Variants: AAC-LC, AAC-LD, AAC-HE, ...

• Vorbis– Second-generation, similar quality to AAC– Open-source, royalty-free (Xiph.Org Foundation)

Future of codecs

• Improving quality– Super-wideband, coding of music– The gap between speech and music codecs is closing– AMR-WB+, G.722.1x moving to music, higher quality– AAC-LD moving to lower delay

• Reducing delay• Increasing robustness

– Shift from bit-error robustness to packet loss robustness

Improved Music Handling

• Background music is poorly handled– Most speech codecs (AMR-NB, G.729, AMR-

WB, Speex etc) are derivatives based on CELP– CELP makes assumptions that are only valid for

speech (and single-note music)– CELP does not perform well on music –

especially at low bit-rate– Music codecs are not suitable for speech

Improved Music Handling

How do you improve the handling of background music?

• Three strategies:1. Increase the bit-rate2. Dual-mode codecs (e.g. AMR-WB+)3. Use non-CELP codecs (AAC-LD, G.722.1x,

G.711.1, CELT, …)

Wideband Extension (WEx) as an interim solution

How do you provide a wideband experience when linking a wideband-capable client to the PSTN?

• Current solution: up-sample the narrowband speech to 16 kHz

• Better solution: Create wideband “artificially” from the narrowband speech

• Support becoming available– WEx capable handsets (Philips for example)– WEx enabled Media Gateway (Vocallo for example)

The Implementationsa.k.a The Role of the Media Gateway

Wideband VoIP DECT - France Telecom

AccessPlatform

IPNetwork

IAD

MobilePlatform

IMS GWTDMNetwork

DLC

AccessPlatform

IAD

Wide Band Extension (WBE)

AccessPlatform

IPNetwork

IAD

IP/DLC

MobilePlatform

IMS GW

TDMNetwork

DLC

AECANR

NLE

LEC

Wide Band ExtensionExpand the signal to create impression of wideband.

WBE

Improving the User Experience

AccessPlatform

IPNetwork

IAD

IP/DLC

MobilePlatform

IMS GW

TDMNetwork

AECANR

NLE

Wideband Lite Acoustic Echo Canceller acts as a complement to badly designed

handset

Wideband Adaptive Noise Reduction reduces noise of mobile handset environment.

Wideband Natural Level Enhancement, uses info from intensity of the voice and SNR to compensate for loud environment of the talker

DLC

The role of the MGW

• When selecting MGW solutions:– Don’t just look for checklist of codecs!– Look for solutions that provide wideband

extension, wideband ECAN, ANR, etc.– Select solutions that incur low latency when

transcoding IP-to-IP communications

Summary

• Clear benefit to the users– Skype changed expectation levels

• Technology enablers already in place– VoIP deployment– CODECS – WB-enabled end-points and MGWs available