Can someone help me understand the bit reservoir system that LAME uses? As I understand it, if the samples in a frame can be encoded with the same accuracy using fewer bits than is required by the bitrate, then this is done, and the extra bits are counted and can be used in future frames where a sample may gain more accuracy from more bits to encode with.

I'm looking for something along the lines of an algorithmic explanation of this process. If anyone can explain, or provide a pointer to some resources, I would much appreciate it.

Thank you!

The MPEG data contains a "data start offset". This can be a negative value which means that the data starts in the middle of the previous frame

The following picture shows an example of an mp3 stream. You can see a frame is divided in three sections: header, side info and main data. the "bit reservoir" is simply unused "main data space" of previous frames.


Here's an older thread with more detailed explanations.

Cheers
SG

Thanks for the info! This is much clearer to me now.