Re: [codec] Codec guidelines and requirements drafts

[Koen Vos <[e-mail]>]

Quoting Jean-Marc Valin:

>> 3) It is also usually stated that VBR is bad for speech transport over
>> the Internet, because it leads to higher loss rates then when CBR is used.
>> [...]
>
> Do you have any reference regarding VBR and loss rate? In any case, I
> think it's clear that both VBR and CBR have valid use cases.

I'm only aware of the opposite relationship between VBR and packet  
loss: short peaks at higher rate are easily handled by the packet  
buffer in the bottleneck (e.g. DSL modem). Even temporarily exceeding  
the network bandwidth only results in a small amount of jitter.
On the other hand, constantly exceeding the network bandwidth not only  
guarantees packet loss, but also keeps the packet buffer filled up,  
easily adding hundreds of milliseconds of extra delay to the call.

Other arguments against using CBR:
- At a given quality level, VBR results in fewer total bits (~ half)  
being sent over the network over the duration of a call.
- VBR helps on the receiving end to estimate the available network  
bandwidth, so that the codec can adjust to it (using a feedback  
channel).

VBR seems better suited for the Internet than CBR.

best,
koen.