Note for 32-bit system users with long audio files

[Alberto Pettarin]

Dear all,

in the unlikely case you are running on a 32-bit system or using a
32-bit Python on a 64-bit system, please remember that no more than 2 GB
of contiguous RAM can be allocated by numpy.

This means that, if you try using aeneas with a very long audio file,
aeneas might fail with an "unexpected exception", which masks a
MemoryError thrown by numpy when trying to bring into memory the WAVE
file converted from your original audio file.

The failure point, in terms of the maximum audio file length, depends on
the aeneas rconf parameters. In theory it is around 4h-long audio files
with default rconf parameters, but it might be much lower, depending on
how much RAM your OS/Python eats.

The failure occurs even if you have plenty of RAM, it is just a
limitation of 32-bit platforms. In the specific case that was brought to
my attention, a user was trying to process a 4h20m long audio file on a
machine with 16 GB of RAM, on a 32-bit Ubuntu. Per se, 4h20m required
about 1.9GB of RAM space, but with OS/Python/numpy overhead, it
triggered the 2GB limit.

Issue logged as: https://github.com/readbeyond/aeneas/issues/172

If you are using a recent machine, you are probably on a 64-bit system,
and you are NOT affected by this.

Best regards,

Alberto Pettarin