An Improved Unsupervised Single-Channel Speech Separation Algorithm for Processing Speech Sensor Signals

Algorithm 1

CNMF+JADE description.

Input: Speech signal dataset, , and O.

1: Initialize each parameter and variable:

 T=, expresses the set of all the pure speech signal data of the speaker waiting to be separated,

 H=, expresses the set of all the mixed speech that is used as the training set,

 O denotes a mixed speech waiting to be separated,

  is a random matrix.

2: whilei < Ndo

3:  The speech data with the identical subscript and from the datasets T and H are selected to train CNMF.

4:  The trained CNMF is employed to separate the mixed speech dataset O to determine and .

5:  The two speech signals acquired from 4 are mixed to obtain a two-channel speech signal and stored in .

6:  A secondary separation is conducted by adopting JADE to obtain and from .

7:   is used as the speech signal to be separated in the next round, and and are removed from the data sets T and H.

8:  Obtain the final separated speech signal.

9:  i = i +1.

10: end while.

Output: All of speaker’s speech signals .