I would like to define a custom numpy
array data type : for some (audio) applications, a np.int24
data type could be useful. A few remarks :
- of course I could use
np.int32
instead and forget about the lost memory space - or I may use some other tricks for seeing an array as a
np.int8
array, and grouping them by 3, but this would involve dirty coding...
... But when using gigabytes of data, the best solution (instead of wasting 1/4 th of a np.int32
) would be to define a new dtype = np.int24
.
Is there a documentation showing how to do that properly ?
Remark : 24 bit
audio is very common in professionnal audio applications, and when dealing with gigabytes of data, using 24 bits array would be very useful.