Expectation-maximisation for speech source separation using convolutive transfer function

CAAI Transactions on Intelligence Technology

article

Xiaofei Li¹  Laurent Girin¹  Radu Horaud¹ 

[1] INRIA Grenoble Rhône-Alpes;GIPSA-lab;Université Grenoble Alpes

关键词: speech enhancement;    expectation-maximisation algorithm;    transfer functions;    speech processing;    Fourier transforms;    microphone arrays;    frequency-domain analysis;    blind source separation;    microphones;    convolution;    source separation;    reverberation;    convolutive transfer function;    under-determined speech source separation;    multichannel microphone signals;    convolutive mixtures;    multiple sources;    time-domain signals;    short-time Fourier;    (STFT) domain;    room filters;    STFT domain;    widely used narrowband assumption;    CTF coefficients;    mixing filters;    STFT coefficients;    expectation-maximisation algorithm;    B0240Z Other topics in statistics;    B6130 Speech and audio signal processing;    B6140 Signal processing and detection;    C1140Z Other topics in statistics;    C5260S Speech processing techniques;   

DOI  :  10.1049/trit.2018.1061

学科分类：数学（综合）

来源: Wiley

【 摘 要 】

This study addresses the problem of under-determined speech source separation from multichannel microphone signals, i.e. the convolutive mixtures of multiple sources. The time-domain signals are first transformed to the short-time Fourier transform (STFT) domain. To represent the room filters in the STFT domain, instead of the widely used narrowband assumption, the authors propose to use a more accurate model, i.e. the convolutive transfer function (CTF). At each frequency band, the CTF coefficients of the mixing filters and the STFT coefficients of the sources are jointly estimated by maximising the likelihood of the microphone signals, which is resolved by an expectation-maximisation algorithm. Experiments show that the proposed method provides very satisfactory performance under highly reverberant environments.

【 授权许可】

CC BY|CC BY-ND|CC BY-NC|CC BY-NC-ND