| 摘要 |
The method involves forming a sum signal (y) and a difference signal (d0) from two microphone signals (x1, x2). Low frequencies in the difference signal are amplified in such way, that the produced signal (d) has the same statistical characteristics as a noise component in the sum signal. A soft decision signal (DS) for a growing probability of an existence of a speech signal is produced as a detector signal, from a ratio of short-term power averages of the sum and difference signals. The method involves using two microphone signals (x1, x2), in which a speech signal component is adjusted through a run-time equalization to provide a speaker alignment. A sum signal (y) and a difference signal (d0) are subsequently formed, as well as derived sizes in an evaluated frequency range. Low frequencies in the difference signal are amplified in such way, that the produced signal (d) has the same statistical characteristics as a noise component in the sum signal. A soft-decision signal (DS) for a growing probability of an existence of a speech signal is produced as a detector signal from a ratio of short-term power averages of the sum- and difference signals.
|
