cVectorOperation

Description

This component performs elementary operations on vectors (i.e. basically everything that does not require history or context, everything that can be performed on single vectors w/o external data (except for constant parameters, etc.))

Fields

  • operation (string) [default: norm]

    A string which specifies the type of operation to perform:

    • norm = normalise vector length (euclidean norm, L2) to 1

    • nr1 = normalise range to +1, -1

    • nr0 = normalise range to +1, 0

    • nma = divide by the maximum absolute value

    • mul = multiply vector by param1

    • add = add param1 to each element

    • log = compute natural logarithm

    • lgA = compute logarithm to base param1

    • nl1 = normalise vector sum (L1 norm) to 1

    • sqrt = compute square root

    • pow = take values to the power of param1

    • exp = raise param1 to the power of the vector elements

    • ee = raise the base e to the power of the vector elements

    • abs = take absolute value of each element

    • agn = add Gaussian noise with mean param1 and std.dev. param2

    • min = take the min of vector and param1

    • max = take the max of vector and param1

    • sum = compute sum of vector elements, there will be a single output only

    • ssm = compute sum of squared vector elements, there will be a single output only

    • ll1 = compute sum of vector elements normalised by the number of vector elements, there will be a single output only

    • ll2 = compute euclidean length (root of sum of squares normalised by vector length), there will be a single output only

    • fla(tten) = flattening of comb filter energy spectra, as in 2007 ICASSP Paper and Ballroom dance style recognition.

    • dBp = convert a power to decibel with 10*log10(x).

    • dBv = convert an amplitude/magnitude/voltage to decibel with 20*log10(x)

    • fconv_aaa_bbb = convert frequency from scale aaa to scale bbb

    • lin = linear (Hz)

    • bark = Bark (Traunmueller, 1990)

    • mel = Mel-scale

    • oct = semitone/octave scale (music), param1 = freq. of first note in Hz.

  • param1 (numeric) [default: 1.0]

    parameter 1

  • param2 (numeric) [default: 1.0]

    parameter 2

  • logfloor (numeric) [default: 1e-12]

    floor for log operation

  • powOnlyPos (numeric) [default: 0.0]

    if 'operation' = 'pow', do not take negative values to the power of 'param1'; instead, output 0. This is necessary to avoid 'nan' values if the exponent is rational.

  • nameBase (string) [default: None]

    base of output feature name when performing n->1 mapping operations (currently 'euc' and 'sum')

  • appendOperationToName (numeric) [default: 0.0]

    (1/0 = yes/no) append the operation name from the 'operation' option to the feature name. This will override any nameAppend option (inherited from cDataProcessor).