Fermé

Voice processing

Speech Detection and Analysis of fundamental frequency Measurements of the logarithm of time and energy short rate transit from the ground a short time, can be used together as the basis of a reliable algorithm for the exact location of authority and end of the voice signal.

For the algorithm to be implemented, we assume that a speaker speaks, without stopping, any part of speech (word, phrase, sentence), during a predetermined intervals and recording the signal for the entire time recording, sampled and stored for further processing. The aim of such a detection algorithm ends, the Finding a reliable authority (B) and end (E) speech, so that any processing and pattern recognition followed, can ignore ambient noise.

The recorded voice signal, s [n], in principle be converted to a typical sampling rate (10 KHz for this algorithm) and then passes from one anoperato filter to eliminate DC component (offset) and the buzz, using an anoperato FIR filter 101 points, constant ripple (equiripple). Then applied processing of short-term energy export and transit rate scratch, using a frame size 40 msec (L = 400 samples for Fs =10000 Hz), sliding frames 10 m sec (R = 100 samples for Fs = 10000 Hz), for basic rate of 100 frames / sec. The two parameters short years, the logarithmic energy (log Er) and the rate of zero crossing (Zr), calculated for each context, r, of the recorded signal. The parameters short years calculated as:

Er = Σΐ=1 (5 [rR+ m ] w [ m]) ,

E r = 10logX0 Er-max (10logw Er),

r

Z =R/(2L )Xm=, 0 \sgn(s [rR+m]) — sgn(s [rR+m — 1 ])|,

where L = 400, R = 100, and w [m] is a window Hamming L points.

We assume that the short years logarithm of the energy, normalized to a maximum value of 0 dB and that Zr is the number of zero crossing at 10msec. It is assumed that the first 100 ms (10 frames) of the recorded signal contain no speech. This is justified in most applications ,due to the large response time of the speakers from the time it is asked to begin to speak for a preset time, and after given the signal for the start of the recording. The mean and standard deviation of the logarithm of energy short time and rhythm crossing from zero, calculated at this early time of 100 m sec, so give us a rough statistical estimate of the background signal. These mean values and standard deviations, and declared as eavg esig logarithm is for energy, and zcavg zcsig rate for zero crossing. Utilizing these measurements, calculated a threshold rate zero crossing, IZCT, as IZCT = max (IF, zcavg + 3 * zcsig).The quantity IF (which takes value 35) is a general threshold for non-detection of non emfonon frames (based on statistical failure rate of transitemfonon sounds). The value of the threshold, IZCT, increases if the background signal, during the first 100 m sec of the recording, showing a high rate transit, as estimated based on the prices and zcavg zcsig. Similarly, for measuring the logarithm of energy, we define a pair thresholds, particularly as ITU, an upper (conservative) threshold, and ITR, a somewhat less conservative threshold for the presence of speech. The ITU and ITR also determined by statistics of the logarithmic energy speech and background signals, and can be changed again, based on conduct that presents the logarithm of the energy in the first 100msec of the recording, according to the relations ITU = constant in the interval [-10 to -20] dB, ITR = max (ITU-10, eavg + 3 * esig). Based on the thresholds, made an initial search ,to find an area in which the curve of the logarithm of energy concentrated around an energy maximum. This is achieved by performing. Search the following procedure:

1. Search for a concentration range of the logarithm of energy(Going forward starting from the frame 1) to find a framework where the logarithm of the energy exceeds the lowest doorstep, ITR. Then check the area around this framework, to make sure that the logarithm of the energy of neighboring frames is above the high threshold, ITU, bef

Compétences : Programmation C++

Voir plus : voice processing, zcavg zcsig, itr max itu eavg esig, xm programming, us algorithm, statistics algorithm, statistical programming, statistical algorithm, starting programming, standard speaker, scratch programming, range search algorithm, programming scratch, programming define, programming 101, pattern search algorithm, pattern finding algorithm, pair programming, pair of points, max programming, lowest range, logarithmic algorithm, first voice recording, first transit, first energy

Concernant l'employeur :
( 0 commentaires ) Greece

N° du projet : #1677886

4 freelance ont fait une offre moyenne de 238 $ pour ce travail

luchenggang

I am a speech preocessing expert. Please check my technology resume in PM.

250 $ USD en 7 jours
(3 Commentaires)
3.4
zoint

I think you want to do speech recognition via academic [url removed, login to view] this is complex algorithm.I have been made a speech recognition in Matlab and C# a long time [url removed, login to view] message back.

300 $ USD en 40 jours
(0 Commentaires)
0.0
matlabCexpert

Hi, I have been working in video image processing since last 8 years. I have exposure in ffmpeg, libmpeg2, mssg kind of decoders/encoders. I know how to play with such tool and its libraries. I have coded such motion Plus

200 $ USD en 2 jours
(0 Commentaires)
0.0
evanov

Hi, please i want to take a moment to read my PM.

200 $ USD en 7 jours
(0 Commentaires)
0.0