Method for vectorial noise-reduction in speech, and implementation device
First Claim
Patent Images
1. A method for vectorial noise-reduction for a source of sound signals that is substantially a point source, comprising the steps of:
- vectorially picking up sound signals by means of at least four equally spaced out microphones positioned in a plane before the source of sound signalsforming linear combinations from the sound signals by additions and subtractions of the sound signals from said microphones;
wherein the sum of all the sound signals from the microphones represents the noisy signal each of the other linear combinations of sound signals comprises as many subtractions as additions of the sound signals and represent noise only signals;
processing the linear combinations of the sound signals from the microphones that have as many subtractions as additions with a vectorial adaptive filter to produce an estimate of the noise of the noisy signal; and
subtracting said estimate of the noise of the noisy signal from the noisy signal.
1 Assignment
0 Petitions
Accused Products
Abstract
A sound pick-up operation is carried out by means of at least four microphones positioned in a plane before the speaker, linear combinations are formed by additions and subtractions of the signals from these microphones and certain of these combinations are used as noise-only sources to process them by an adaptive vector filter to produce an estimation of the noise that is subtracted from the noisy signal.
-
Citations
9 Claims
-
1. A method for vectorial noise-reduction for a source of sound signals that is substantially a point source, comprising the steps of:
-
vectorially picking up sound signals by means of at least four equally spaced out microphones positioned in a plane before the source of sound signals forming linear combinations from the sound signals by additions and subtractions of the sound signals from said microphones; wherein the sum of all the sound signals from the microphones represents the noisy signal each of the other linear combinations of sound signals comprises as many subtractions as additions of the sound signals and represent noise only signals; processing the linear combinations of the sound signals from the microphones that have as many subtractions as additions with a vectorial adaptive filter to produce an estimate of the noise of the noisy signal; and subtracting said estimate of the noise of the noisy signal from the noisy signal. - View Dependent Claims (2, 3, 4, 5, 6, 7)
-
- 8. A method for vectorial noise-reduction in a source that is substantially a point source, comprising at least one set of at least eight equally spaced out microphones, each followed by an amplifier, a sampling device, the sampling devices being all synchronous and connected to a binary conversion matrix comprising serializing connections and inverters, a first one of these outputs corresponding to the serializing of all the inputs, each of these other outputs corresponding to a different combination of of serializing operations and input inversions, the first output of this combiner device being connected to the (+) input of a subtractor and the others being connected via an adaptive filter to the (-) input of the subtractor.
Specification