Kurvengeräusche detektion

##algorithm

• squeal 1102 1976 Hz width < 10Hz minimal duratuion 40-60 ms beginning abrupth -phase?

• flanging brodband beginning abrupt ending slow phase?

A. snkg =(livello totale -(flanging)livello filtrato lp 1500) smooth(snKg) (not simmetric (past!)) kg if(snkg > delta)

B.

##Algorithm evaluation specifity: true negative / false positive sensityve: true positive / N positive events ##Kurvengerüsche package

• measuredValues :
• time signals:
• dsp : Calculate the squeal and flanging of a set of microphones given a set of signal during a passby
• vizualise Widget

import hand handle mbbm measured and evaluated values in tables store kg processed values export data frames for further calculations

handle the time signals of a given passby ID

Sqeal noise of a passby is a time function $z(t) \in { 0,1}$. 0 is no squeal 1 is squeal. This function can be discretized in a time sequence $z_n = z(t_n)$ with $t_n = n \Delta t$. This is a given deterministic sequence.

We are interested in the quantity $T_z = \int z(t) dt = \sum z_n \Delta t$ of a given passby.

We have a detection algorithm (measurement of squeal noise) which has an output $M_n \in {0,1}$. The Algorithm is not perfect, $M_n$ is random variable. The algorithm/ test of squel noise is characterized by:

• true positive rate (sensitivity): $P(M_n=1|z_n = 1) = p_{tp}({z_n})$
• false positive rate: $P(M_n=1|z_n = 0) = 1 - p_{tn}({z_n})$
• false segative rate: $P(M_n=0|z_n = 1) = 1 - p_{tp}({z_n})$
• true negative rate(Specifity): $P(M_n=0|z_n = 0) = p_{tn}({z_n})$

$M$ is a test to know if a person is allergic. $z$ is the allergic state of a person. The tests has following charachteristics:

• true positive rate : $P(M=1|z = 1) = 0.8$
• false positive rate: $P(M=1|z = 0) = 0.1$
• false negative rate: $P(M=0|z = 1) = 0.2$
• true negative rate: $P(M=0|z = 0) = 0.9$

if a people has a positive test $M=1$ how is probable is that he is allergic? given that the true probability of allergic persons is $P(z=1)= pz = 0.01$ and $P(z=0)=1-pz = 0.99$

• step 1: test probability outcome $P(M)$ on $z$: $P(M=1) = P(M=1|z = 1)\cdot P(z=1) + P(M=1|z = 0)\cdot P(z=0) = 0.8 \cdot 0.01 + 0.1\cdot 0.99$

pMz1 = c(0.8,0.1)
pMz0 = c(0.2,0.9)
pz = c(0.01,0.99)
pM1 = sum(pMz1*pz)
print(pM1)

$P(M=0) = 0.2 \cdot 0.01 + 0.9\cdot 0.99$

pM0 = sum(pMz0*pz)
print(pM0)

• step 2: The question anzwers is given by $$ P( z = 1 | M=1) = \frac{P( M = 1| z = 1 ) \cdot P(z=1)}{ P(M = 1)}$$

  pz1M1 = 0.8*0.01/pM1 
  print(pz1M1)
  

given a test and many outcomes, estimate $pz$

#Selecting the intervals : The script run_CaseCreatorWidget.py call the GUI to select intervals where one hears flanging or squealing noise. Make sure there are the following files in the same path :

• caseToAnalyze.json : Contains the mID to be analysed. Each entry looks like the following:

"m_00403_4" : {
		"case" : {
			"location" : "Biel",
			"mID" : "m_00403",
			"measurement" : "Vormessung",
			"mic" : 4,
			"Te" : 19.0,
			"Tb" : -2.375,
			"author" : null
		},
		"wavPath" : "wav\\m_00403_mic_4.wav",
		"plotData" : {
			"LAfast" : [[-8.0, -7.875, ... , 53.8153076171875, 53.974822998046875]]
		},
		"tmax" : 21.203105926513672,
		"tmin" : -7.999987602233887
	}

• AppCS folder containing the info.html and info.css files for the information page. It also contains the icon, the images for the html page, and so on.

• firstcases.json (optional) : contains the first cases to evaluate from the caseToAnalyze list.

#Analysing the algorithms To analyse how an algorithms selects the intervals and how it compares with authors selected cases, you can use run_AdminAlgorithmWidgets.py and select from Files/new Create new Admin.

#Building the exe file We use cx_freeze package to build the executable file. Setup.py is the file to be called with cx_freeze using the following command (on Windows): pathtopython\python cxsetup.py build

This will create a folder called build aside the directory KG. You need to do the following in order to fix some cx_freeze issues (for points 1 and 2, the files are also available in the folder missingFilesforExe):

1. Copy the file _ufuncs.pyd into build folder (you will find it under pathtopython/lib/site-packages/scipy/special
2. Copy the files libifcoremd.dll and libmmd.dll into build folder (you will find them under pathtopython/lib/site-packages/numpy/core
3. Move the folder dateutils from the zipped build/library into build (i.e. remove it from the archive and place it into build folder)
4. Add the file caseToAnalyse.json in build
5. Finaly add the corresponding wav files inside build/wav

First install distutils. To create the package run python setup.py build. Then to install the package use pip : pip install kg.