opensmile_emobase_live4.conf

///////////////////////////////////////////////////////////////////////////////////////
///////// > openSMILE configuration file for live emotion recognition < ///////////////
/////////   base set of 988 features, 1st level functionals          //////////////////
/////////   of low-level descriptors such as MFCC, Pitch, LSP, ...   //////////////////
/////////                                                            //////////////////
/////////  * written 2009 by Florian Eyben *                         //////////////////
/////////                                                            //////////////////
///////// (c) 2014 audEERING UG (haftungsbeschränkt),                //////////////////
/////////     All rights reserverd.                                  //////////////////
///////////////////////////////////////////////////////////////////////////////////////


///////////////////////////////////////////////////////////////////////////////////////
;
; This section is always required in openSMILE configuration files
;   it configures the componentManager and gives a list of all components which are to be loaded
; The order in which the components are listed should match 
;   the order of the data flow for most efficient processing
;
///////////////////////////////////////////////////////////////////////////////////////
[componentInstances:cComponentManager]
 ;this line configures the default data memory:
instance[dataMemory].type=cDataMemory
instance[waveSource].type=cRawWaveSource
instance[fr25].type=cFramer
 ;; VAD (energy-based)
instance[energy].type=cEnergy
instance[turn].type=cTurnDetector
 ;;; 40 ms frames features:
instance[fr40].type=cFramer
instance[w40].type=cWindower
instance[fft40].type=cTransformFFT
instance[fftmagphase40].type=cFFTmagphase
instance[acf40].type=cAcf
instance[cepstrum40].type=cAcf
 ; Pitch...
instance[pitchACF].type=cPitchACF
 ;;; 25 ms frames features:
instance[pe].type=cVectorPreemphasis
instance[win].type=cWindower
instance[fft].type=cTransformFFT
instance[fftmagphase].type=cFFTmagphase
instance[mspec].type=cMelspec
 ; MFCC
instance[mfcc].type=cMfcc
instance[lpc].type=cLpc
 ; Line Spectral Frequencies
instance[lsp].type=cLsp
 ; Zero-Crossings
instance[mzcr].type=cMZcr
 ; Intensity and Loudness (narrow-band approximation)
instance[intens].type=cIntensity
 ;;; all LLD concattenated and smoothed using a moving average filter
instance[lld].type=cContourSmoother
 ; delta coefficients of LLD
instance[delta1].type=cDeltaRegression
 ;;; functionals over FULL input (e.g. turns)
instance[functL1].type=cFunctionals
instance[dataSink].type=cRawDataSink

;; run single threaded (nThreads=1)
; NOTE: a single thread is more efficient for processing small files, since multi-threaded processing involves more 
;       overhead during startup, which will make the system slower in the end
nThreads=1
;; do not show any internal dataMemory level settings 
; (if you want to see them set the value to 1, 2, 3, or 4, depending on the amount of detail you wish)
printLevelStats=0


/////////////////////////////////////////////////////////////////////////////////////////////
/////////////////////////   component configuration  ////////////////////////////////////////
/////////////////////////////////////////////////////////////////////////////////////////////
; the following sections configure the components listed above
; a help on configuration parameters can be obtained with 
;  SMILExtract -H
; or
;  SMILExtract -H configTypeName (= componentTypeName)
/////////////////////////////////////////////////////////////////////////////////////////////

[waveSource:cRawWaveSource]
writer.dmLevel=wave
monoMixdown = 1
blocksize_sec = 0.01
channels = 2
sampleRate = 16000

[fr40:cFramer]
reader.dmLevel=wave
writer.dmLevel=frames40
frameSize = 0.040
frameStep = 0.010
frameMode = fixed
frameCenterSpecial = center

[w40:cWindower]
reader.dmLevel=frames40
writer.dmLevel=win40frame
copyInputName = 1
processArrayFields = 1
winFunc = ham
gain = 1.0
offset = 0

[fft40:cTransformFFT]
reader.dmLevel=win40frame
writer.dmLevel=fftc40
copyInputName = 1
processArrayFields = 1
inverse = 0
 ; for compatibility with 2.2.0 and older versions
zeroPadSymmetric = 0

[fftmagphase40:cFFTmagphase]
reader.dmLevel=fftc40
writer.dmLevel=fftmag40
copyInputName = 1
processArrayFields = 1
inverse = 0
magnitude = 1
phase = 0

[acf40:cAcf]
reader.dmLevel=fftmag40
writer.dmLevel=acf40
nameAppend = acf
copyInputName = 1
processArrayFields = 1
usePower = 1
cepstrum = 0
acfCepsNormOutput = 0

[cepstrum40:cAcf]
reader.dmLevel=fftmag40
writer.dmLevel=cepstrum40
nameAppend = acf
copyInputName = 1
processArrayFields = 1
usePower = 1
cepstrum = 1
acfCepsNormOutput = 0
oldCompatCepstrum = 1
absCepstrum = 1

[pitchACF:cPitchACF]
  ; the pitchACF component must ALWAYS read from acf AND cepstrum in the given order!
reader.dmLevel=acf40;cepstrum40
writer.dmLevel=pitch
processArrayFields=0
maxPitch = 500
voiceProb = 1
voiceQual = 0
HNR = 0
F0 = 1
F0raw = 0
F0env = 1
voicingCutoff = 0.550000

[fr25:cFramer]
reader.dmLevel=wave
writer.dmLevel=frames
frameSize = 0.025
frameStep = 0.010
frameMode = fixed
frameCenterSpecial = center


[pe:cVectorPreemphasis]
reader.dmLevel=frames
writer.dmLevel=framespe
copyInputName = 1
processArrayFields = 1
k=0.97

[win:cWindower]
reader.dmLevel=framespe
writer.dmLevel=winframe
copyInputName = 1
processArrayFields = 1
winFunc = ham
gain = 1.0
offset = 0

[fft:cTransformFFT]
reader.dmLevel=winframe
writer.dmLevel=fftc
copyInputName = 1
processArrayFields = 1
inverse = 0
 ; for compatibility with 2.2.0 and older versions
zeroPadSymmetric = 0

[fftmagphase:cFFTmagphase]
reader.dmLevel=fftc
writer.dmLevel=fftmag
copyInputName = 1
processArrayFields = 1
inverse = 0
magnitude = 1
phase = 0

[mspec:cMelspec]
reader.dmLevel=fftmag
writer.dmLevel=mspec1
copyInputName = 1
processArrayFields = 1
htkcompatible = 1
nBands = 26
lofreq = 0
hifreq = 8000
usePower = 1
inverse = 0
specScale = mel

[mfcc:cMfcc]
reader.dmLevel=mspec1
writer.dmLevel=mfcc1
copyInputName = 0
processArrayFields = 1
firstMfcc = 1
lastMfcc =  12
cepLifter = 22.0
htkcompatible = 1

[lpc:cLpc]
reader.dmLevel=framespe
writer.dmLevel=lpc
copyInputName = 1
processArrayFields = 1
method = acf
p = 8
saveLPCoeff = 1
lpGain = 0
saveRefCoeff = 0
residual = 0
forwardFilter = 0
lpSpectrum = 0

[lsp:cLsp]
reader.dmLevel=lpc
writer.dmLevel=lsp
copyInputName = 1
processArrayFields = 0

[intens:cIntensity]
reader.dmLevel=frames
writer.dmLevel=intens
copyInputName = 1
processArrayFields = 1
intensity=1
loudness =1

[mzcr:cMZcr]
reader.dmLevel=frames
writer.dmLevel=mzcr
copyInputName = 1
processArrayFields = 1
zcr = 1
amax = 0
mcr = 0
maxmin = 0
dc = 0

[lld:cContourSmoother]
reader.dmLevel=intens;mfcc1;lsp;mzcr;pitch
writer.dmLevel=lld
writer.levelconf.nT = 2500
writer.levelconf.isRb=1
writer.levelconf.growDyn=0
nameAppend = sma
copyInputName = 1
noPostEOIprocessing = 0
smaWin = 3


// ---- delta regression of LLD ----
[delta1:cDeltaRegression]
reader.dmLevel=lld
writer.dmLevel=lld_de
blocksize=1
writer.levelconf.nT = 2500
writer.levelconf.isRb=1
writer.levelconf.growDyn=0
nameAppend = de
copyInputName = 1
noPostEOIprocessing = 0
deltawin=2

[energy:cEnergy]
reader.dmLevel=winframe
writer.dmLevel=energy
rms=1
log=0

// the voice activity detection (turn detector)
[turn:cTurnDetector]
reader.dmLevel=energy
writer.dmLevel=isTurn
 ; overwrite data in output level, if it has not been read and level is full
 ; we must set this here, since the level isTurn is a dead-end in our configuration
writer.levelconf.noHang=1
 ; send a message to the functionals component at turn end
messageRecp = functL1
 ; send turn start / turn end event messages to the debug turn wave file output component
eventRecp = turnDump
  ;; you may not increase this over the writer.levelconf.nT value of the lld and lld_de levels!
  ;; so by default not more than 24!
maxTurnLength = 12
maxTurnLengthGrace = 3
idx = 0
nPost = 30
nPre = 10
useRMS = 1
 ; disable auto adjustment of VAD threshold, this does not work well yet.
autoThreshold = 0
 ; instead, set the threshold manually to a default value.
 ; this derived from the RMS energy from normalised sample values (0..1)
 ; --> !!!!! you might need to adjust this value depending on your recording setup !!!!!!! <-------------------
threshold = 0.0015
 ; --> !!!!! you might need to adjust this value depending on your recording setup !!!!!!! <-------------------


// statistical functionals
[functL1:cFunctionals]
reader.dmLevel=lld;lld_de
writer.dmLevel=func
copyInputName = 1
 ; frameMode = var will enable the functionals component to listen for messages from the turn detector
frameMode = var
functionalsEnabled=Extremes;Regression;Moments;Percentiles
Extremes.max = 1
Extremes.min = 1
Extremes.range = 1
Extremes.maxpos = 1
Extremes.minpos = 1
Extremes.amean = 1
Extremes.maxameandist = 0
Extremes.minameandist = 0
 ; Note: the much better way to normalise the times of maxpos and minpos
 ; is 'turn', however for compatibility with old files the default 'frame' 
 ; is kept here:
Extremes.norm = frame
Regression.linregc1 = 1
Regression.linregc2 = 1
Regression.linregerrA = 1
Regression.linregerrQ = 1
Regression.qregc1 = 0
Regression.qregc2 = 0
Regression.qregc3 = 0
Regression.qregerrA = 0
Regression.qregerrQ = 0
Regression.centroid = 0
Regression.doRatioLimit = 0
Regression.centroidRatioLimit = 0
Moments.doRatioLimit = 0
Moments.variance = 0
Moments.stddev = 1
Moments.skewness = 1
Moments.kurtosis = 1
Moments.amean = 0
Percentiles.quartiles = 1
Percentiles.iqr = 1

  //////////////////////////////////////////////////////////////////////
 ///////////////////  data output configuration  //////////////////////
//////////////////////////////////////////////////////////////////////

// ----- you can use this to save the features extracted  ------
[dataSink:cRawDataSink]
reader.dmLevel=func
fieldInfo = 1
//////---------------------- END -------------------------///////