Matlab Audio Analysis Library

The current document is an outline of the Matlab Audio Analysis Library which accompanies the book Introduction to Audio Analysis:A MATLAB® Approach, 1st Edition.

The provided material is organized as follows::

• Folder "library"
• In the root of this folder you can find:
  • the core m-files of the Matlab Audio Analysis Library
  • a number of mat files
• Folder "demos" contains m-files used for demonstrating particular functionalities of the library. Most of these demos are presented in the book. Note, that, in order to run the demos, one has to add the root path (i.e. the path of the "library" folder) in the MATLAB path.
• Folder "data" contains basic audio data that have been used to evaluate and train several algorithms described in the book.

Contents of "/data"

Contents of "/library"

In the following table we provide a short description of the core Matlab functions of the library, i.e the functions stored in the "library" folder (not the ones stored in the "demos" folder).
For a description of the five (5) .mat files (i.e the kNN models of the respective classification tasks), please refer to Table 5.1 of the book.

m-file	Description	Chapter
audioRecorderOnline	Demonstrates the audio recording using the audiorecorder() MATLAB function. Calls the audioRecorderTimerCallback() callback function.	2
audioRecorderTimerCallback	Callback function used to record audio data (through the audiorecorder() MATLAB function)	2
classifyKNN_D_Multi	Classifies an unknown sample using the kNN algorithm, in its multi-class mode. Returns probability estimates	5, 6
computePerformanceMeasures	Computes the confusion matrix and performance measures of a classification process	5
dctCompress	Demonstrates the use of DCT for audio compressing	3
dctDecompress	Demonstrates the use of DCT for audio (de)compressing	3
dynamicTimeWarpingItakura	Computes the Dynamic Time Warping cost between two feature sequences based on the Itakura local path constraints	7
dynamicTimeWarpingSakoeChiba	Computes the Dynamic Time Warping cost between two feature sequences based on the Sakoe-Chiba local path constraints	7
em_alg_function	EM algorithm for estimating the parameters of a mixture of normal distributions, with diagonal covariance matrices	7
EM_pdf_est	EM estimation of the pdfs of c classes	7
evaluateClassifier	Implements the repeated hold out and leave-one-out validation methods	5
feature_chroma_vector	Computes the chroma vector of a short-term window	4
feature_energy	Computes the energy of a shortterm window	4
feature_energy_entropy	Computes the entropy of energy of a short-term window	4
featureExtractionDir	Extracts mid-term features for a list of WAV files stored in a given folder	8
featureExtractionFile	Reads a WAVE file and computes audio feature statistics on a mid-term basis	4,5,6
feature_harmonic	Computes the harmonic ratio and fundamental frequency of a window (autocorrelation method)	4
feature_mfccs	Computes the MFCCs of a short-term window (Based on Slaney'sAuditory Toolbox)	4
feature_mfccs_init	Initializes the computation of the MFCCs (see feature mfccs())	4
feature_spectral_centroid	Computes the spectral centroid of a short-term window	4
feature_spectral_entropy	Computes the spectral entropy of a short-term window	4
feature_spectral_flux	Computes the spectral flux of a short-term window	4
feature_spectral_rolloff	Computes the spectral rolloff of a short-term window	4
feature_zcr	Computes the zero crossing rate of a short-term window	4
fftExample	Demonstrates how to use the getDFT() function	3
fftSNR	Demonstrates the use of the getDFT() function using a noisy signal	3
fileClassification	Demonstrates the classification of an audio segment from a WAVE file (not to be confused with mtFileClassification() which performs joint segmentation-classification of an audio file	5
fld	Finds a linear discriminant subspace using the LDA algorithm. Used for dimensionality reduction in the context of music visualization. This m-file has not be implemented by the authors, but it was taken from Mathworks File Exchange, Fisher Linear Discriminant Analysis, by Sergios Petridis	8
getDFT	Returns the (normalized) magnitude of the DFT of a signal.	3,4
kNN_model_add_class	Adds an audio class to a kNN classification setup. As the kNN classifier requires no actual training, the function it only performs a feature extraction stage for a set of WAVE files, stored in a given directory	5
kNN_model_load	Loads a kNN classification setup, i.e., a feature matrix for each class, along with the respective normalization parameters (means and standard deviations of the features)	5
mixturepdf	Computes the value of a pdf that is given as a mixture of normal distributions, at a given point.	7
mp3toWav	Performs MP3 to WAVE conversion with the FFMPEG command-line tool	2
mp3toWavDIR	Transcodes each MP3s of a given folder to the WAVE format, using the FFMPEG command-line tool	2
mtFeatureExtraction	Computes the mid-term statistics for a set of sequences of short-term features. It returns a matrix, whose columns contain the vectors of mid-term feature statistics.	4
mtFileClassification	Splits an audio signal into fixedsize segments and classifies each segment separately (fixed-size window segmentation)	5,6
musicMeterTempoInduction	Performs joint estimation of the music meter and tempo of a music recording	8
musicThumbnailing	Extracts pairs of thumbnails from music recordings	8
musicVisualizationDemo	Demonstrates three linear dimensionality reduction methods for music content visualization (random projection, PCA and LDA)	8
musicVisualizationDemoSOM	Demonstrates SOM-based music content visualization	8
plotFeaturesFile	Plots a given feature sequence that has been computed over a WAVE file	4
printPerformanceMeasures	Prints a table of classification performance measures (confusion matrix, recall, etc) in LATEX format	5
readWavFile	Demonstrates how to read the contents of a WAVE file, using two different modes: (a) all the contents of the WAVE file are loaded (b) blocks of data are read and each block is processed separately	2
readWavFileScript	Generates experiments that measure the elapsed time of different WAVE file I/O approaches	2
scaledBaumWelchContObs	Implements the scaled version of the Baum-Welch algorithm (continuous features)	7
scaledBaumWelchDisObs	Implements the scaled version of the Baum-Welch algorithm (discrete observations)	7
scaledViterbiContObs	Implements the Viterbi algorithm for continuous features	7
scaledViterbiDisObs	Implements the Viterbi algorithm for discrete observations	7
scriptClassificationPerformance	Loads a kNN classification setup (stored in a mat file) and extracts the respective classification performance measures. For the best value of k , it prints the respective confusion matrix and class-specific performance measures.	5
segmentationCompareResults	Visualizes two different segmentation results for the sake of comparison.	6
segmentationPlotResults	Provides a simple user interface to view and listen to the results of a segmentation - classification procedure.	6
segmentationProbSeq	Segments an audio stream based on the estimated posterior probabilities for each class. Implements: (a) naive merging and (b) viterbi-based probability smoothing. To be called after mtFileClassification().	6
segmentationSignalChange	Basic unsupervised signal change segmentation (no classifier needed).	6
showHistogramFeatures	This auxiliary function is used to plot the histograms of a particular feature for different audio classes. It has been used to generate the histograms of Chapter 4	4
silenceDetectorUtterance	Computes the endpoints of a single speech utterance. Based on Rabiner and Schafer, Theory and Applications of Digital Speech Processing, Section 10.3.	6
silenceRemoval	Applies a semi-supervised algorithm for detecting speech segments (removing silence) in an audio stream stored in a WAVE file.	6
smithWaterman	Implements the Smith-Waterman algorithm for sequence alignment	7
soundOS	An alternative to the Matlab sound() function, in case problems are encountered in Linuxbased systems	2
speakerDiarization	Implements a simple unsupervised speaker diarization procedure.	6
stFeatureExtraction	Breaks an audio signal to possibly overlapping short-term windows and computes sequences of audio features. It returns a matrix whose rows correspond to the extracted feature sequences	4
stpFile	Demonstrates the short-term processing stage of an audio signal.	2
viterbiBestPath	Finds the most-likely state sequence given a matrix of probability estimations. Used for smoothing segmentation results.	6
viterbiTrainingDo	Implements the Viterbi training scheme for the case of discrete observations	7
viterbiTrainingMultiCo	Implements the Viterbi training scheme for the case of continuous, multidimensional features, under the assumption that the density function at each state is Gaussian	7
viterbiTrainingMultiCoMix	Implements the Viterbi training scheme for the case of Gaussian mixtures	7

M-files dependency graph