Closed boeddeker closed 4 years ago
This PR allows reducing the memory consumption for wpe_v8. The loopy code already has a low memory footprint, but the code produces a copy of the input.
wpe_v8
For example in CHiME-6 https://github.com/kaldi-asr/kaldi/blob/6247d3cf9fb50f2e7dbd9faefd21175d04435177/egs/chime6/s5_track1/local/run_wpe.py#L58 the input has a size of 2.13 GiB and is never reused after the dereverberation. The inplace option allows saving an array, that has the same size as the input array.
2.13 GiB
This PR allows reducing the memory consumption for
wpe_v8
. The loopy code already has a low memory footprint, but the code produces a copy of the input.For example in CHiME-6 https://github.com/kaldi-asr/kaldi/blob/6247d3cf9fb50f2e7dbd9faefd21175d04435177/egs/chime6/s5_track1/local/run_wpe.py#L58 the input has a size of
2.13 GiB
and is never reused after the dereverberation. The inplace option allows saving an array, that has the same size as the input array.