AcousticFeatures

This package implements a set of methods to compute generic acoustic features in AxisArrays for time-series acoustic data.

The acoustic features are:

Acoustic Feature	Constructor
Energy	Energy
Myriad	Myriad
Frequency Contours	FrequencyContours
Sound Pressure Level	SoundPressureLevel
Impulse Statistics (number of impulses, mean and variance of inter-impulse intervals)	ImpulseStats
Symmetric Alpha Stable Statistics (α and scale)	SymmetricAlphaStableStats
Entropy (temporal entropy, spectral entropy and entropy index)	Entropy
Zero Crossing Rate	ZeroCrossingRate
Spectral Centroid	SpectralCentroid
Spectral Flatness	SpectralFlatness
Permutation Entropy	PermutationEntropy
PSD	PSD
Acoustic Complexity Index	AcousticComplexityIndex
Statistical Complexity	StatisticalComplexity
Acoustic Diversity Index	AcousticDiversityIndex

Installation

using Pkg; pkg"add https://github.com/ymtoo/AcousticFeatures.jl.git"

Usage

using AcousticFeatures, DSP, SignalAnalysis, Plots

function compare(sc1, sc2)
    plot(sc1.axes[1] ./ fs, vec(sc1.data),
         xlabel = "Time (sec)",
         ylabel = "Permutation Entropy",
         label  = "without chirp",
         color  = :blue,
         dpi    = 150,
         thickness_scaling = 1.0,
         legend=:bottomleft
    )
    plot!(sc2.axes[1] ./ fs, vec(sc2.data),
          xlabel = "Time (sec)",
          ylabel = "Permutation Entropy",
          label  = "with chirp",
          color  = :red,
          dpi    = 150,
          thickness_scaling = 1.0,
          legend=:bottomleft
    )
end

N  = 2400
fs = 2400
v  = randn(Float64, 3*N)
s  = real(chirp(500, 1000, 1.0, fs))
x  = copy(v); 
x[N:2*N-1] += s

specgram(signal(x,fs); fs=fs,nfft=128)

winlen = 2400
noverlap = 1200
pe = PermutationEntropy(5, 1, true, true)
sc1 = Score(
    pe,
    v;
    winlen = winlen,
    noverlap = noverlap
)
sc2 = Score(
    pe,
    x;
    winlen = winlen,
    noverlap = noverlap
)
compare(sc1, sc2)

N  = 2400
fs = 2400
v  = randn(Float64, 3*N)
Ns = fs ÷ 100
bpf = fir(13, 500, 1000; fs=fs) 
s  = 10 .* filtfilt(bpf, randn(Float64, Ns)) #real(chirp(500, 1100, 0.1, fs))
x  = copy(v); 
Ngap = fs ÷ 10
x[N:N+Ns-1] += s
x[N+5Ngap:N+5Ngap+Ns-1] += s
x[N+9Ngap:N+9Ngap+Ns-1] += s

specgram(signal(x,fs); nfft=128, fs=fs)

winlen = 2400
noverlap = 1200
pe = PermutationEntropy(5, 1, true, true)
sc1 = Score(
    pe,
    v;
    winlen = winlen,
    noverlap = noverlap,
)
sc2 = Score(
    pe,
    x;
    winlen = winlen,
    noverlap = noverlap,
)
compare(sc1, sc2)