[Matroska-devel] RFC 80085 - Required TimecodeScale for sample accurate seeking

[Steve Lhomme]

Now let's see another example. A file using only audio (MKA file).
You only have one audio track at 44100 Hz.

My point is that the Global TimecodeScale can remain at 1,000,000. But 
the Track TimecodeScale should be 1000/44100 = 0,022676...

So imagine you have Audio Blocks of 1152 samples. Here is a list Of 
values in the first Cluster :
Block #   Timecode 1st Smpl   Final Timecode   Block Timecode

    1             0                  0                0
    2         1152/44100 s       26,122,449ns       1152
    3        2*1152/44100 s      52,244,898ns       2304
    4        3*1152/44100 s      78,367,347ns       3456
......

(using formula AudioTimecode = int16 * float * uint64)

So for each Block you know exactly the sample count on which starts the 
Block.

This way, if you want to seek to sample 2353 you compute the timecode :
AudioTimecode = 2353/44100 = 53,356,009ns

So you know that the Block to use is the Block # 3 in Cluster # 1. And 
with the sample count, you know what is the exact sample you're looking 
for (the 50th).

I don't know if the computing of the Block Timecode might lead to 
rounding errors...