why OverSampledDFTAnalysisBank using a backward fft? about distant_speech_recognition HOT 2 CLOSED

The details of filter bank design is described here:
https://www.dropbox.com/s/2hc32piufzvm3ea/filter-bank-design.pdf?dl=0

The short paper is also published in ICASSP2008:
"Filter bank design based on minimization of individual aliasing terms for minimum mutual information subband adaptive beamforming", ICASSP 2008.

from distant_speech_recognition.

The details of filter bank design is described here:
https://www.dropbox.com/s/2hc32piufzvm3ea/filter-bank-design.pdf?dl=0

The short paper is also published in ICASSP2008:
"Filter bank design based on minimization of individual aliasing terms for minimum mutual information subband adaptive beamforming", ICASSP 2008.

The details of filter bank design is described here:
https://www.dropbox.com/s/2hc32piufzvm3ea/filter-bank-design.pdf?dl=0

The short paper is also published in ICASSP2008:
"Filter bank design based on minimization of individual aliasing terms for minimum mutual information subband adaptive beamforming", ICASSP 2008.

thanks kkumatani! I made a mistake when opening this issue, I did not make a full description about my question, the complete issue is here https://github.com/kkumatani/distant_speech_recognition/issues/9, the first link you provide can not be opened, I looked through the second link, but It seems did not explain why a backward fft(inverse fft) adopted in analysis filter bank but not a forward fft, maybe I should research more about it

from distant_speech_recognition.