Re: [codec] WG Review: Internet Wideband Audio Codec (codec)

[Gregory Maxwell <[e-mail]>]

Richard Shockey [[e-mail]] wrote:
>>  I can see the motivation to pay big bucks for video codecs. Using
>>  Mpeg4 can reduce your bandwidth costs and save real money. I can see
>>  why there was a big incentive to save money on audio codecs in the
>>  1990s.
>>  At this point an audio codec is going to have to save a huge amount ot
>>  bandwidth to be worth the hassle, let alone the cost of using
>>  encumbered technology.
> Its not about the bandwidth. Its about the quality of the voice in
> occasionally lossy networks landline or mobile.

"full quality" mono audio takes around 44.1/16bit linear, you can argue that a little higher or lower is required for full transparency in some conditions or another, but 44.1k/16 is what CDDA provides. It's a good number and commonly available on hardware today.

As uncompressed thats 705kbit/sec before you get into packetization overhead.  Thats per-communication channel, each way. A good perceptual lossy codec can get you down to under 100kbit/sec while preserving transparency for most material, with most listeners, most of the time. 

Networks have improved a lot. It actually is viable to send uncompressed CDDA across many wide-area networks today, at least in small amounts,  but I think we're a long way from when compression doesn't have considerable advantages, even for audio.  I think it would be fair to say that networks have become fast enough other considerations such as error robustness, perceived-transparency,  computational cost, latency, licensing considerations, etc.  are now often more important than the absolute minimization of bandwidth. but savings on the order of 8-12 : 1 are not something which can be ignored. 

8-12:1 is still a material difference in capacity and cost.   Audio is cheaper now relative to the network, but that just means we can do more of it: More quality, more capacity.

Moreover, every bit wasted on uncompressed audio could instead be spent on additional redundancy.   In a bandwidth plentiful model I'd much rather run a 128kbit/sec/ch codec sending each packet three times than a 705kbit/sec lossless stream.