Speech extraction method based on multiple reference signals ICA algorithm

In order to improve the signal noise ratio (SNR) and signal interference ratio (SIR) of the output speech signal and extract pure target speech in the environment with strong noises and interferences, a novel target speech extraction method based on multiple reference signals independent component analysis (ICA) algorithm is proposed in this paper. The reference signals are acquired by source localization, beamforming and wavelet translation algorithms. Then the target speech is estimated by FastICA algorithm based on the negative entropy combined with the reference signals. Simulations and experiments using microphone array signals demonstrated that background noises and interference speech were reduced and pure target speech waveform and spectrogram were recovered effectively. Compared to the traditional beamforming and ICA algorithms, the proposed method achieved better performance such as higher SNR and SIR.