cVectorMVN

Description

This component extends the base class cVectorTransform and implements mean/variance normalisation. You can use this component to perform on-line cepstral mean normalisation. See cFullinputMean for off-line cepstral mean normalisation.

Fields

  • nameAppend (string) [default: None]

    A string suffix to append to the input field names (default: empty)

  • meanEnable (numeric) [default: 1.0]

    1 = enable normalisation to 0 mean

  • stdEnable (numeric) [default: 1.0]

    1 = enable standardisation to stddev 1

  • normEnable (numeric) [default: 0.0]

    1 = enable normalisation (scaling) to range -1 to +1, if meanEnable=1 (x-mean)/((max-min)*0.5), or to range 0..1 if meanEnable=0 (x-min)/(max-min) ; (this can NOT be used in conjunction with stdEnable)

  • minMaxNormEnable (numeric) [default: 0.0]

    1 = enable normalisation (scaling) to range -1 to +1 with only min and max values: x=2*(x-min)/(max-min)-1

  • spectralFlooring (numeric) [default: 0.0]

    1 = enable spectral subtraction by flooring all spectral bins below the mean (to the value of 'specFloor') and not touching bins above the mean (except if you set subtractMeans=1). (works only if meanEnable=1 is the only normalisation option set)

  • subtractMeans (numeric) [default: 0.0]

    (only relevant if spectralFlooring=1) : 1 = if input values are above the mean, subtract the mean (else floor to 'specFloor').

  • specFloor (numeric) [default: 1e-10]

    The value to which bins will be set that are floored.

  • htkLogEnorm (numeric) [default: 0.0]

    1 = enable HTK compatible log energy normalisation (this also sets normEnable=1, meanEnable=0)