Professional Documents
Culture Documents
Audio Networking: Bob Moses AES Vice President Western Region USA/Canada
Audio Networking: Bob Moses AES Vice President Western Region USA/Canada
Presentation Outline
Why Network Audio? How? Ramifications Listening Tests
Engineering Perspective
Digital Audio systems benefit by being networked Share resources
Lowers cost Increases performance
S to ra g e
N e tw o rk Access
A u d io R e n d e re r A u d io R ip p e r U ser In te rfa c e
A u d io O u tp u ts A u d io I/O A u d io In p u ts
CAT UTP
100 m
100 m
100 m
no
STP / Coax
500 m
100 m
25 m
no
Multi fibre
mode
2 km
550 m
300 m
25 km
20 km
5 km
40 km
Physical Constraints
1394b CAT-5 POF HPCF GOF STP Distanc e 100m 50m 100m 100m
4.5m
100 Mbps X X X X X
200 Mbps
400 Mbps
800 Mbps
1.6 Gbps
3.2 Gbps
* * X X X X X X X X X X
Heterogeneous Networks
No network is perfect, and no single network will win universal adoption Heterogeneous networks are comprised of two or more network technologies Bridges and gateways are crucial devices for connecting subnets
Bridges connect two or more networks running the same protocol stack (e.g. HomePNA to Ethernet) Gateways connect networks that use different protocols (e.g. Ethernet to 1394)
Ramifications
Systems evolve from autonomous devices to communities of devices. Control becomes decentralized and migrates to the edge devices. This is a revolutionary change (not evolutionary)
Paradigm Shifts
Once there is too much information to own, we must evolve methods to access and organize it. Todays systems are equipment-centric, and very complex. Tomorrows systems need to be content-centric, and very simple. The distinction between ownership, and renting, and a service is already blurred in the video industry. Thats likely to happen in the music industry as well. Duplicating and distributing plastic disks is barbaric.
Content-Based Searching
Chris Weare & Ted Tanner: Breakthroughs in machine learning, pattern recognition, and feature extraction can be applied to audio to characterize the similarity between two recordings. A neural network was trained with data from over 115,000 songs to produce a commercial system that provides over 90% success. Humans no longer need to listen to music because computers can be trained to listen for them!
Content Protection
Copy inhibit bits (SCMS) Watermarking Fingerprinting SDMI DTCP
Do any of these technologies truly work? Should we create technical solutions to legal problems, or legal solutions to technical problems?
Metadata
Elizabeth Cohen: We generate terabytes of new music data each day The machine and format I record on today wont be around in 20 years. Its important to preserve the music experience itself, not just the bitstream How do you record and regenerate an experience? Some of our audio treasures might be lost when future devices are unable to play them back. Elizabeth Cohen recommends that if a device leaves a footprint on the audio, the content must describe it via metadata.
Dont forget about the Creative Team During Encoding! Bob Clearmountain: Producing content for the Internet is not merely a file transfer process creative decisions must be made in the mastering process that ultimately affect the listening experience. The process of compressing audio for the low bandwidth of the Internet is very similar to the old days of squeezing music into a plastic groove of a vinyl LP during mastering, and that the compromises involved should be a creative process involving the creative team. Internet mix should also be considered in addition to the CD, radio, extended dance mix, and other mixes.
Compression Formats
Required for asynchronous networks MPEG-1 Layer 3 (MP3) MPEG 2, 4, MPEG 7 AAC Windows Media Real Audio/Video Quicktime Ogg Vorbis VQF Beatnik etc.
Listening Tests
Edward II: Wicked Men Track #7: Shes Gone to California Original (44.1kHz, 16-bit) Compressed Formats
MP3 WMA 7.0 Real Audio 6.0 (AAC can not be played back due to copy protection glitch)
160Kbps
Random ordering
64Kbps
Random ordering