What is the JSTK (and what not)

The Java speech toolkit (JSTK) provides a native implementation of both library/API and applications for speech recognition, speaker verification, speech visualization (including transcription tools), and evaluation of related human rater tasks. JSTK is not a full-blown LVCSR toolkit with pre-trained models, sorry.

Where is JSTK used?

• Hidden Markov Model-Based Smart Annotation for Benchmark Cyclic Activity Recognition Database Using Wearables. C.F. Martindale, S. Sprager and B.M. Eskofier, Sensors 2019, 19(8), 1820.
• Digital sports: http://www5.cs.fau.de/our-team/jensen-ulf/projects/drop-jump-phase-length-calculation-with-hidden-markov-models/
• Automatic Intelligibility Assessment of Speakers After Laryngeal Cancer by Means of Acoustic Modeling T Bocklet, K Riedhammer, E Nöth, U Eysholdt, T Haderlein. Journal of Voice 26 (3), 390-397, 2012
• Erlangen-CLP: A Large Annotated Corpus of Speech from Children with Cleft Lip and Palate T Bocklet, A Maier, K Riedhammer, U Eysholdt, E Nöth. Language Resources and Evaluation (LREC) 2014
• LMELECTURES: A Multimedia Corpus of Academic Spoken English. K Riedhammer, M Gropp, T Bocklet, F Hönig, E Nöth, S Steidl, SLAM@ INTERSPEECH, 2013.
• Interactive Approaches to Video Lecture Assessment. K Riedhammer. Logos Verlag Berlin GmbH, 2012.
• A Software Kit for Automatic Voice Descrambling. K Riedhammer, M Ring, E Nöth, D Kolb. IEEE Workshop on Security and Forensics in Communication Systems, 2012.
• Compensation of extrinsic variability in speaker verification systems on simulated Skype and HF channel data K Riedhammer, T Bocklet, E Nöth. Acoustics, Speech and Signal Processing (ICASSP), 2011.
• Drink and Speak: On the Automatic Classification of Alcohol Intoxication by Acoustic, Prosodic and Text-Based Features. T Bocklet, K Riedhammer, E Nöth. INTERSPEECH, 2011.
• Java Visual Speech Components for Rapid Application Development of GUI Based Speech Processing Applications. S Steidl, K Riedhammer, T Bocklet, F Hönig, E Nöth. INTERSPEECH, 2011.
• Demo system for age recognition: http://www.nacht-der-wissenschaften.de/2009/ansicht_sammel.php?id=234#234

Building JSTK

You can import JSTK as a project to Eclipse, or use the provided gradle file to build the complete jar file (and dependencies) in build/libs/{jstk-latest,libs/*}.jar

Using JSTK

The easiest way to use the command line tools is to make a script like the following:

$ cat jstk
#!/bin/bash

CLASSPATH=/home/sikoried/Work/workspace/jstk/bin:/home/sikoried/Work/lib/jtransforms-2.3.jar:/home/sikoried/Work/lib/Jama-1.0.2.jar:/home/sikoried/Work/lib/FJama.jar:/home/sikoried/Work/lib/log4j-1.2.16.jar

java -Xmx7G de.fau.cs.jstk.$@

Can be used then as

$ ./jstk app.Decoder

Maintenance and Support

The Java Speech Tooklit (JSTK) was originally developed and maintained by the Speech Group at the University of Erlangen-Nuremberg. It is designed to provide both API and applications for the most popular speech tasks such as speech recognition, speaker verification, speech transcription and annotation and evaluation of human rater tasks.

The JSTK is licensed under GPLv3 and welcomes any contributions in terms of extensions, bugfixes, feature requests or comments. Though this toolkit is intended to allow full-scale speech applications, special APIs such as the Google Android Speech API or Java Speech API are not implemented (but may however be implemented in the future).

Contributions are welcome -- please create issues or PRs.

Dependencies (managed through gradle)

JTransforms - http://sites.google.com/site/piotrwendykier/software/jtransforms

JTransforms provides (discrete) signal processing routines such as FFT, DHT, and DCT. It relies on state-of-the-art algorithms and even allows multi- threading.

JAMA - http://math.nist.gov/javanumerics/jama/

The JAMA package provides linear algebra routines such as decompositions (e.g. Eigen, LR, QR) and matrix operations (calculus, inversions). A single precision copy is maintained at the Speech Group named FJAMA, and a snapshot can be found in the download section of this project.

log4j - http://logging.apache.org/log4j

log4j provides scalable logging capabilities, however, JSTK only uses the very basic features of it -- for now. Currently, we use v 1.2.16

jspeex - http://sourceforge.net/projects/jspeex/files/jspeex

jspeex is used for decoding speex-compressed (*.spx) files. speex is a lossy codec comparable to MP3 or Ogg Vorbis, but optimized for speech, see http://www.speex.org. Currently using version 0.9.7.

junit - http://www.junit.org

Testing framework for java. currently used only sporadically. Currently using version 4.