1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
|
Readme file for Oxygen beamforming source code distribution
-----------------------------------------------------------
This is a very very beta distribution of beamforming source code from
MIT LCS.
There is only one source file, main.c, and one header file,
main.h. You can compile everything using the Makefile included in the
distribution.
The input to the program is a text file containing floating
point values for the signal read on each of the microphones. For n
microphones, each line of the text files represents a temporal sample
and should contain n floating point values separated by spaces, e.g.:
-1.8569790e-004 -9.0919049e-004 3.6711283e-004 -1.0073081e-005 ...
There are several modes of operation for the program:
1. The most basic mode is to process the microphone data and to
calculate the output based on one beam focused on a particular point
in space. The coordinates for the microphones and the focus point are
specified inside main.c (eventually to be moved to a separate file).
2. Far field search mode. This mode assumes a far-field source (which
means we have a planar wavefront) and a linear array, and sweeps over
180 degrees in the plane of the array. The microphone coordinates are
specified in main.c, and the NUM_ANGLES constant defines how many
angle values should be tested (a value of 180 means one beam per each
degree). The energy of the signal over a particular window
(ANGLE_ENERGY_WINDOW_SIZE) is computed for each beam. The direction
with maximum energy is considered the direction that the speech signal
is coming from, and is printed out by the program.
3. Near-field hill climbing mode. This mode accepts a starting
coordinate and attempts to "hill-climb" through the space seeking the
maximum energy. Each of the x, y, and z coordinates are perturbed in
the positive and negative directions at each time interval
(GRID_ENERGY_WINDOW_SIZE) by a step size (GRID_STEP_SIZE). This
perturbation, along with the original coordinate, produces seven
coordinates to be tested. The direction with the maximum energy
replaces the current reference coordinate. For instance, if we have a
starting reference coordinate of (1,1,0) and our step size is 0.01, we
will evaluate the energy for the following seven beams:
(1,1,0)
(0.99,1,0)
(1.01,1,0)
(1,0.99,0)
(1,1.01,0)
(1,1,-0.01)
(1,1,0.01)
Now let's say the beam (1,1.01,0) has the maximum energy; then this
coordinate will replace the original reference coordinate of (1,1,0).
For methods 2, and 3, we are not outputting anything to disk, we are
just printing the result. This is because we have just started to work
with these methods, and have not applied them in real systems. This
code is currently being ported to RAW.
To get a list of parameters for the delay_and_sum executable that is
generated when the source is compiled, just type ./delay_and_sum .
There is some sample data included with the program, in the data
directory. There is some data for a near-field and far-field
source. The README.txt file in each directory specifies the microphone
and source position. The data1 file, when processed with a beamformer
aligned in the proper direction should produce something like a sinc
function (see
http://ccrma-www.stanford.edu/~jos/Interpolation/sinc_function.html).
The data2 file should produce an audio signal of a woman saying "the
simplest method". If the beamformer is aligned properly, the noise
should be reduced significantly over the source signal from only one
of the microphones (use print_datafile.pl to isolate one
microphone). You can convert the data file that the program produces
to wave files using sox.
---------------------------------
Eugene Weinstein
ecoder@mit.edu
|
