Computes and displays a chord diagram of a set of audio files or of a set segments extracted from a single audio file.
scd(input, f, sl, wl = 512, wn = "hanning", ovlp = 0, flim = NULL,
rmoffset = TRUE, threshold = NULL, HCA = TRUE,
grid.col = terrain.colors, names, plot = TRUE, verbose = TRUE, ...)THe function returns a list of two items:
spectral similarity matrix
the classification result of the HCA, if HCA is TRUE
a character vector, either (i) the path to the directory where
.wav files are stored, (ii) directly the names of the .wav files to be
processed, or (iii) a single .wav file to be segmented by the
duration set with the argument sl.
sampling frequency of wave (in Hz). Does not need to
be specified if embedded in the .wav files contained in input.
segment length in duration if input is a single file.
window length for the analysis spectral (even number of points) (by default = 512).
window name for the spectral analysis, see
ftwindow (by default "hanning").
overlap between two successive windows (in %) for the spectral analysis.
a numeric vector of length 2 to select a frequency band (in kHz).
a logical to sepcify whether DC offset should be
removed. By default TRUE.
a numeric value in ]0,1[ to be applied to the similarity distance. All similairty distances below this threshold will not depicted.
logical, if TRUE the sectors are colored according
to clusters automatically obtained with a hierarchical cluster analysis (HCA).
name of color palette to color the sectors and
the links). By default terrain.colors.
names of the sectors, if empty then the names of the .wav files or the time stamps of the segments.
logical, if TRUE plots the chord diagram.
logical, if TRUE prints the computation progress
of the mean spectra.
other chordDiagram parameters.
Adèle de Baudouin, Jérôme Sueur
The soundscape chord diagram (SCD) aims at representing similarities
between audio files or audio segments extracted from a single audio file. The
mean frequency spectrum of each file/segment is computed using a
STFT. These frequency spectra are then (1) pairwised compared using a
similarity distance (see function diffcumspec, and (2)
automatically clustered with a hierarchical cluster analysis (HCA) (see function
HCPC of FactoMiner). The
resulting similarity matrix is then given as an input to the function
chordDiagram. The width of the sectors and the links are
based on the spectral similarity matrix. The color of the sectors and the links
follow the HCA classification.
de Baudouin, A, Couprie P, Michaud F, Haupert S, Sueur J -- Similarity visualization of nature and music soundscapes, in prep.
diffcumspec