API

abstract type Features end

Abstract type of features extractors.

Base.summary(f::Features)

Display a table summarizing each field of the given feature (name, value, type).

Only available in HTML context.

Base.summary(c::ComposedFunction{<:Any,<:Features})

Provides a summary for each feature

AudioSources.load(s::AbstractAudioSource, withprops::Bool; subrange::Union{AbstractRange, Colon}=:, ch=1)

Similar to load(s::AbstractAudioSource), with aditional option: set withprops to true to return a FeaturesProperties object with the audio data.

FeaturesProperties{T}

T is used to know which feature is associated.

Fields

• fs is the frequency sampling of the input to the feature extractor
• fs_init keep track of the initial fs when several Features are chained
• scale is a Vector giving the feature resolution, in its given unit

DataProps{T}::DataType = Tuple{AbstractArray, FeaturesProperties{T}}

Alias type that is the input and output type of all Features functions.

Base.summary(props::FeaturesProperties)

Display a table summarizing the given features properties.

Frames <: Features

Segment resulting from framing the signals. This representation is usually used to extract the short-term Fourier transform.

Frames(; <keyword arguments>)

Initialize frames with default values if not specified.

Arguments

• frameduration = 0.025 in seconds
• framestep = 0.01 time between two frames in seconds
• dithering = 0.0 add gaussian noise to the signal
• preemph = 0.97 improve signal-to-noise ratio by boosting high frequencies
• removedc = true
• windowname = "hann" framing window, one of ["hann", "hamming", "povey", "rectangular"]
• padding = 0 amount of padding
• dropedge = true

(f::Frames)((x, props)::DataProps{AudioSources.AbstractAudioSource})

Apply Frames to the given signal.

Result is a matrix of size (frame length, number of frames), and the new FeaturesProperties

STFT <: Features

Short-Term Fourier Transform.

(f::STFT)((X, props)::DataProps{Frames})

Apply Short-Term Fourier Transform to the given Frames matrix.

Result is a matrix of size (≈ frame length / 2, number of frames), and the new FeaturesProperties.

FBANK <: Features

Mel spectrum.

FBANK(; <keyword arguments>)

Initialize FBANK with default values if not specified.

Arguments

• numfilters = 26 number of filters ("triangles")
• lofreq = 80 lowest frequency to keep
• hifreq = -400 Nyquist frequency + hifreq is the highest frequency to keep

(f::FBANK)((X, props)::DataProps{STFT})

Apply Mel filterbank to the given matrix.

Result is a matrix of size (numfilters, number of frames), and the new FeaturesProperties.

MFCC <: Features

Mel Frequency Cepstral Coefficients.

MFCC(; <keyword arguments>)

Initialize MFCC with default values if not specified.

Arguments

• nceps = 13 number of cepstral coefficients
• liftering = 22 lifter value

(f::MFCC)((X, props)::Union{DataProps{FBANK}, DataProps{STFT}})

Apply MFCC to the given matrix.

Result is a matrix of size (nceps, number of frames), and the new FeaturesProperties.

Autocorr <: Features

Autocorrelation

(f::Autocorr)((X, props)::DataProps{Frames})

Apply autocorrelation to the given frames.

Result is a matrix of size (frame length, number of frames), and the new FeaturesProperties

AddDeltas <: Features

Add the derivatives to the features.

AddDeltas(; <keyword arguments>)

Initialize AddDeltas with default values if not specified.

Arguments

• order = 2 derivative order
• winlen = 2 length of delta window

(f::AddDeltas)((X, props)::DataProps{<:Features})

Apply AddDeltas to the given Features matrix.

Result is a matrix of size (nb input matrix rows * (order+1), number of frames), and the new FeaturesProperties