TarsosDSP

TarsosDSP is a Java library for audio processing. Its aim is to provide an easy-to-use interface to practical music processing algorithms implemented, as simply as possible, in pure Java and without any other external dependencies. The library tries to hit the sweet spot between being capable enough to get real tasks done but compact and simple enough to serve as a demonstration on how DSP algorithms works.

TarsosDSP features an implementation of a percussion onset detector and a number of pitch detection algorithms: YIN, the Mcleod Pitch method and a “Dynamic Wavelet Algorithm Pitch Tracking” algorithm. Also included is a Goertzel DTMF(Dual tone multi frequency) decoding algorithm, a time stretch algorithm (WSOLA), resampling, filters, simple synthesis, some audio effects, and a pitch shifting algorithm.

To show the capabilities of the library, TarsosDSP example applications are available. Head over to the TarsosDSP release directory for freshly baked binaries and code smell free (that is the goal anyway), oven-fresh sources.

Quickly Getting Started with TarsosDSP

The source code of TarsosDSP is compatible with Java 11.

Add this to your build.gradle file:

repositories {
    maven {
        name = "TarsosDSP repository"
        url = "https://mvn.0110.be/releases"
    }
}
dependencies {
    implementation 'be.tarsos.dsp:core:2.5'
    implementation 'be.tarsos.dsp:jvm:2.5'
}

Add this to your pom.xml file:

<repository>
  <id>be.0110.repo-releases</id>
  <name>0110.be repository</name>
  <url>https://mvn.0110.be/releases</url>
</repository>
<dependency>
  <groupId>be.tarsos.dsp</groupId>
  <artifactId>core</artifactId>
  <version>2.5</version>
</dependency>
<dependency>
  <groupId>be.tarsos.dsp</groupId>
  <artifactId>jvm</artifactId>
  <version>2.5</version>
</dependency>

Citing TarsosDSP

Some information about TarsosDSP can be found in the paper TarsosDSP, a Real-Time Audio Processing Framework in Java, by Joren Six, Olmo Cornelis, and Marc Leman, in Proceedings of the 53rd AES Conference (AES 53rd), 2014. If you use TarsosDSP in academic research, please cite this paper.

@inproceedings{six2014tarsosdsp,
  author      = {Joren Six and Olmo Cornelis and Marc Leman},
  title       = {{TarsosDSP, a Real-Time Audio Processing Framework in Java}},
  booktitle   = {{Proceedings of the 53rd AES Conference (AES 53rd)}}, 
  year        =  2014
}

Building TarsosDSP with Gradle

To build TarsosDSP from source the gradlew script should get you started. The following commands fetch the source and builds the library:

git clone --depth 1 https://[email protected]/JorenSix/TarsosDSP.git
cd TarsosDSP
./gradlew build

Source Code Organization & Developing

The library is separated into three module folders: 1) the main core functionality is found in core/src/main/java, TarsosDSP example applications are found in examples/src/main/java, JVM audio I/O in jvm/examples/src/main.

TarsosDSP Examples

To see TarsosDSP in action thre are a number of examples. These examples have either a graphical user interface or a command line interface. To see a window in which you can start all GUI examples the jar file needs to be started without arguments:

gradle shadowJar
java -jar examples/build/libs/examples-all.jar

To use the command line examples, execute the jar with the example name as first argument. For example java -jar examples/build/libs/examples-all.jar feature_extractor pitch audio.mp3 starts the feature_extractor application to extract pitch from an audio file.

To get a list of all command line examples: java -jar examples/build/libs/examples-all.jar list

Credits

TarsosDSP was developed at University College Ghent, School of Arts between 2009 and 2013, from late 2013 the project is supported by University Ghent, IPEM.

The TarsosDSP borrows algorithms from various other libraries or research paper. Below a complete list of credits can be found.

• The onset detector implementation is based on a VAMP plugin example by Chris Cannam at Queen Mary University, London. The method is described in Drum Source Separation using Percussive Feature Detection and Spectral Modulation by Dan Barry, Derry Fitzgerald, Eugene Coyle and Bob Lawlor, ISSC 2005.
• For the implementation of the YIN pitch tracking algorithm. Both the the YIN paper and the GPL’d aubio implementation were used as a reference. Matthias Mauch (of Queen Mary University, London) kindly provided the FastYin implementation which uses an FFT to calculate the difference function, it makes the algorithm up to 3 times faster.
• The Average Magnitude Difference (AMDF) pitch estimation algorithm is implemented by Eder Souza and adapted for TarsosDSP by myself.
• For the MPM pitch tracking algorithm, the paper titled A Smarter Way To Find Pitch by Philip McLeod and Geoff Wyvill was used.
• The Dynamic Wavlet pitch estimation algorithm is described in Real-Time Time-Domain Pitch Tracking Using Wavelets by Eric Larson and Ross Maddox. The implementation within TarsosDSP is based on the implementation in the Dynamic Wavelet Algorithm Pitch Tracking library by Antoine Schmitt, which is released under the MIT open source license, a license compatible with the GPL.
• The audio time stretching algorithm is described in An Overlap-Add Technique Based on Waveform Similarity (WSOLA) For Hight Quality Time-Scale Modifications of speech by Werner Verhelst and Marc Roelands. As a reference implementation the WSOLA implementation by Olli Parviainen in the LGPL SoundTouch – an open-source audio processing library was used.
• The FFT implementation used within TarsosDSP is by Piotr Wendykier and is included in his GPL’d JTransforms library. JTransforms is the first, open source, multithreaded FFT library written in pure Java.
• The sample rate conversion feature is implemented by Laszlo systems in the GPL’d libresample4j library. libresample4j is a Java port of Dominic Mazzoni’s libresample 0.1.3, which is in turn based on Julius Smith’s Resample 1.7 library
• Various FFT window functions are done by Damien Di Fede and Corban Brook for the GPL’d Minim project.
• Beat induction based on onsets and saliences is done using code from Simon Dixon’s BeatRoot system software is licensed under the GPL. The algorithm is documented in the 2001 JNMR paper Automatic Extraction of Tempo and Beat From Expressive Performances and in the 2007 JNMR article Evaluation of the Audio Beat Tracking System BeatRoot
• A complex domain onset detection function is implemented using Aubio as an inspiration. Aubio, by Paul Brossiers contains very clean object oriented c-code, the cleanest c-code I have ever seen. The algorithm is described in Complex Domain Onset Detection for Musical Signals by Christopher Duxbury, Mike E. Davies, and Mark B. Sandler, in Proceedings of the Digital Audio Effects Conference, DAFx-03, pages 90-93, London, UK, 2003
• An implementation of the Constant-Q transform by Karl Helgason for the GPL’d RasmusDSP project has been adapted for use in TarsosDSP. More information about the Constant-Q transform can be found in the following papers Calculation of a Constant Q Spectral Transform by Judith C. Brown, An Efficient Algorithm for the Calculation of a Constant Q Transform, by Judith C. Brown and Miller S. Puckette, and The Constant Q Transform by Benjamin Blankertz
• The Haar Wavelet Transform implemented in TarsosDSP is based on the pseudocode found in “Wavelets Made Easy” by Yves Nievergelt.
• The lifting scheme wavelet package is developed by Ian Kaplan and is based on the concepts explained in Ripples in Mathematics: The Discrete Wavelet Transform by A. Jensen and A. la Cour-Harbo, Springer. It is only slightly modified for easy use in TarsosDSP
• The frequency domain pitch shifter is developed by Stephan M. Bernsee and is based on the concepts explained in Pitch Shifting Using The Fourier Transform. For the moment it is a rather direct translation of the c implementation. It was released under a ’ Wide Open License’. As the name of the license suggests, it is a license withouth much restrictions.
• The granulator is developed for the beads project. It was released under a GPL-license.

Changelog

First release which includes several pitch trackers and a time stretching algorithm, amongst other things. Downloads and javadoc API can be found at the TarsosDSP release directory

Changed how the audio dispatcher stops. Added StopAudioProcessor.
Added FastYin implementation by Matthias Mauch
Added AMDF pitch estimator by Eder Souza

Modified the interface of PitchDetector to return a more elaborate result structure with pitch, probability and a boolean “is pitched”.
Added an implementation of an envelope follower or envelope detector.

TarsosDSP can do audio synthesis now. The first simple unit generators are included in the library.
It has a new audio feature extraction feature, implemented in the FeatureExtractor example.
Added ASCII-art to the source code (this is the main TarsosDSP 1.3 feature).

Included a resample feature, implemented by libresample4j. Together with the WSOLA implementation, it can be used for pitch shifting (similar to Phase Vocoding). A pitch shifting example (both with a CLI and a UI) is added in the 1.4 version of the TarsosDSP library as well.

Converted TarsosDSP to maven. This is known as the Malaryta-release. The “Malaryta” release is provided to you by RikkiMongoose (idea, documents, git things) and Ultar (converting to maven, refactoring). Malaryta is the capital of Malaryta Raion, Brest Region in the Republic of Belarus. Both of developers spent their childhood in Brest, and think that title Malaryta is as strange as Ubuntu or Whistler. The 1.5 release also includes various FFT window functions from the cool Minim project by Damien Di Fede.

This release features practical onset and beat detection algorithms. A complex domain onset detection and a spectral flux onset detection algorithm are added. This release also includes a way to guess a beat from onsets. Parts of the BeatRoot system, by Simon Dixon, are included to this end. Also included in this release is an implementation of the Constant-Q transform.

This release adds the ability to extract the MFCC from an audio signal. Also an example of the Constant-Q transform is added, together with a reusable visualization class library. The build system is reverted back to pure ANT

With this release it is possible to extract spectral peaks from an FFT and get precise frequency estimates using phase info. An example application called SpectralPeaks is added as well.

This release includes a Haar Wavelet Transform and an example of an audio compression algorithm based on Haar Wavelets. It also includes a significant change in package naming.

The 2.0 version is worth the major version update since it offers out-of-the-box support for Android. The release has no more dependencies on the parts of the java runtime that are not included in Android. To offer this support some packages have been shifted around. The code that does I/O is dependent on the runtime (JVM or Dalvik) and is abstracted using the be.tarsos.dsp.io package.

The 2.1 version restructures some of the source files. All source can now be found in src. The ant build file is adapted to reflect this change. This version also includes an STFT pitch shifter. There was already a time domain pitch shifter included and now a frequency domain implementation is present as well.

The 2.2 version includes a new AudioDispatcher. It has been reviewed toroughly and now behaves predictably for the first and last buffers as well. To prevent compatibility issues the version has been changed.

The 2.3 version includes improved Android support: audio decoding on Android can be done using a provided, statically compiled ffmpeg binary. The ffmpeg decoding functionality for JVM has been improved as well. If no ffmpeg executable is found it is downloaded automatically from here.

Some small changes to the pipe decoder. Now it is possible to set a start and duration for the incoming decoded audio from ffmpeg.

Changes to the build system: moved from ant to gradle. Better documentation, testing and CI with github actions. Release of maven packages.