DMCA
Soft mask methods for single-channel speaker separation (2007)
| Venue: | IEEE Trans. Audio, Speech, Lang. Process |
| Citations: | 31 - 2 self |
BibTeX
@ARTICLE{Reddy07softmask,
author = {Aarthi M. Reddy and Bhiksha Raj},
title = {Soft mask methods for single-channel speaker separation},
journal = {IEEE Trans. Audio, Speech, Lang. Process},
year = {2007},
pages = {1766--1776}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Abstract—The problem of single-channel speaker separation at-tempts to extract a speech signal uttered by the speaker of interest from a signal containing a mixture of acoustic signals. Most algo-rithms that deal with this problem are based on masking, wherein unreliable frequency components from the mixed signal spectro-gram are suppressed, and the reliable components are inverted to obtain the speech signal from speaker of interest. Most current techniques estimate this mask in a binary fashion, resulting in a hard mask. In this paper, we present two techniques to separate out the speech signal of the speaker of interest from a mixture of speech signals. One technique estimates all the spectral compo-nents of the desired speaker. The second technique estimates a soft mask that weights the frequency subbands of the mixed signal. In both cases, the speech signal of the speaker of interest is recon-structed from the complete spectral descriptions obtained. In their native form, these algorithms are computationally expensive. We also present fast factored approximations to the algorithms. Ex-periments reveal that the proposed algorithms can result in sig-nificant enhancement of individual speakers in mixed recordings, consistently achieving better performance than that obtained with hard binary masks. Index Terms—Signal separation, soft masks, speaker separation. I.
Keyphrases
speech signal soft mask method single-channel speaker separation soft mask unreliable frequency component factored approximation acoustic signal speaker separation native form mixed signal reliable component mixed recording hard mask binary fashion current technique mixed signal spectro-gram desired speaker second technique single-channel speaker separation at-tempts sig-nificant enhancement individual speaker index term signal separation frequency subbands complete spectral description hard binary mask spectral compo-nents
