Programmer Guide/Command Reference/EVAL/aseg1

This function implements a simple automatic signal segmentation based on the energy (rms) track of the signal.

---

Usage 1

aseg1(x, 0, n, m, xmin, omax)

x

data vector (usally rms track in dB)

n

minimum length of a signal segment in frames; 0 < n < nrow(x)

m

smoothing length; 0 <= m; if m is greater than zero, the segmentation is performed on a smoothed data vector xs.

xs[i] = avr(x[i-m], .., x[i+m])

xmin

segment threshhold

omax

offset for segment center; 0 < omax

Description

1. A segment is a continuous range of x, where all values are greater than xmin
2. The center of a segment is a continuous range of values inside a segment, where all values are greater than xmax-omax (with: xmax is the maximum value inside the segment).

Result 1

A matrix r with 4 rows. Each column of r defines one segment. Note that all values of r are frame indices (or row indices) of the data vector x.

r[0,i]	beginning of the segment i
r[1,i]	beginning of the center of the segment i
r[2,i]	end of the center of the segment i
r[3,i]	end of the segment i

with: i=0..nrow(r)

---

Usage 2

fft(x)

x

signal vector or matrix; if x is a matrix a spectrum of each column is computed

Result 2

A matrix y with ncol(x) columns and L+2 rows, where each column y[*,j] contains the complex spectrum of the column (channel) x[*,j]. The transformation length L is set to npow2(nrow(x)).

---

Usage 3

fft(x, n {, ytype, poffset, prange, aref})

x

signal vector or matrix; if x is a matrix a spectrum of each column is computed

n

desired length of analysis window;

• If n < nrow(x), the analysis window length L is set to nrow(x), otherwise L is set to n.
• If the analysis window length L is a power of 2 (L=2^M), the fft algorithm is used, otherwise the dft is used.
• If L is greater than nrow(x), zero padding is applied to the signal.

ytype

select the type and format of the computed spectrum (default=0) -> see Result 3

poffset

offset in samples to the signal begin or the selected zero phase position (default=0)

poffset=0	-> phase[i] = atan2(im[i], re[i])
otherwise	-> phase[i] = (atan2(im[i], re[i]) - 2*pi*i/L * poffset) % (2 * pi)

prange

selects the range of phase values (default=0)

prange=0	-> 0 <= phase[i] < 2*pi
otherwise	-> -pi <= phase[i] < pi

aref

reference amplitude if the log. spectrum (ytype=4) is requested (default=1)

Result 3

A matrix y with ncol(x) columns, where each column y[*,j] contains the spectrum of the column (channel) x[*,j]. The type and the length of the spectra is selected by the argument ytype.

ytype	description	content of y[*,j]	nrow(y)
0	complex spectrum in cartesian format	{ re0, im0, re1, im1, ... }	L+2
1	complex spectrum in polar format	{ amp0, phase0, amp1, phase1, ... } with: ampi=sqrt(rei2 + imi2), phasei see poffset	L+2
2	amplitude spectrum	{ amp0, amp1, ... }	L+1
3	power spectrum	{ amp02, amp12, ... }	L+1
4	logarithmic amplitude spectrum	{ lev0, lev1, ... } with: levi=20*log10(ampi/aref)	L+1

---

See also

ifft, dft, dct, cepstrum, lpc, complex arithmetic

<function list>

aseg1

Automatically segment a signal. Apply automatic segmentation to the vector x.

Usage:

aseg1(x, xmin, omin, omax, nmin, navr)

Parameters:

x

A data vector (e.g. an rms function).

xmin

The pause threshold.

omin

The offset for segment detection. E.g. x[i] > xmin+omin.

omax

The offset for the segment center detection. E.g. x[i] > max("segment")-omax.

nmin

The minimum segment duration in frames. A segment is assigned if the segment condition is true for at least nmin frames.

navr

The averaging length. If navr is greater than 0 then all elements x[i] are replaced by avr(x[i-navr .. i+navr]).

Result:

A matrix (or vector) y with 4 rows and one column for each detected segment. Note that all values of y are frame indices.