format long g
myRecording = randn(5,1)
wavebinary = dec2bin( typecast( single(myRecording(:)), 'uint16'), 16 ) - '0';
size(wavebinary)
Notice that is N x 16 where N is number of samples times 2 (you are converting myRecording to single no matter whether it is integer or single or double, so the result of the single() will be 32 bits per sample no matter what the original bits per sample. Then you typecast() into two 16 bit halves, and you convert each 16 bit value to binary. Each converted value creates a 16 bit row.
Now look further and see that you construct
sounddata = double(bin2dec( char( reshape( wavebinary, 16, []).'+'0')));
So you would take that 10 x 16 data in which the rows represent the binary of half of one sample, and you would reshape it to be 16 rows... which in this case would give you 10 columns. But that is the wrong bit order. You are taking like
1 : 16
17 : 32
33 : 48
and rearranging it like
[1 17 33 ... 16 32 48 ...]
The flaw here is that when you create wavebinary you should be doing
wavebinary = reshape((dec2bin( typecast( single(myRecording(:)), 'uint16'), 16 ) - '0').', 1, []);
and now with the transpose the bit order would follow the way you are reconstructing.
