[FFmpeg-devel] Deterministic Noise Generation option for audio codecs that involve noise generation (RNG)

[Reimar Döffinger]

On 07.09.2013, at 21:02, Jonathan Campbell <jonathan at castus.tv> wrote:

>> The way I understand this, this means that a file that contains
>> the same AC-3 data over and over will then generate a constantly
>> repeating noise pattern.
>> That sounds like it might be worse than just switching of noise
>> generation, at least in some cases.
>> If possible it should at least get a frame number as input, then
>> it is at least possible to avoid that two consecutive frames generate
>> exactly the same noise, though admittedly an application might have a hard
>> time always knowing the exact frame number.
> Huh, what exactly would motivate an AC-3 decoder to generate the exact same frame data over and over again, other than pure silence, which would not involve any noise coding anyway?

Maybe none of the AC3 encoders would do this, but many speech codecs have a "comfort noise" mode, where they do exactly encode that: a frame with silence but a bit of generated noise.
So maybe it's overkill of AC3 but as an API it would be nice if it worked well for these cases, too.