SIPAC

Signals and Data

Identification, Processing,
Analysis, and Classification
Framework for Mass Data Processing with
Modules for Data Storage, Production
and Configuration

SIPAC key features

SIPAC is a framework for automated and interactive processing of signal and data files. Processing tasks
are user definable by taking advantage of splitting the task in sub modules, which are combined in so called
signal flows. The system is scalable with respect to processing power and processing software. Data and
result files produced by the SIPAC Processing Server can be stored in the SIPAC Archive Server for later
use.
SIPAC offers a pre-defined set of processing and analysis modules and organizes the handling of separate
software solutions, in particular those of different suppliers, running on machines with different operating systems. An expert for each of the modules defines, configures and administrates all necessary elements of the
implementation on the SIPAC platform. After this step, the module can easily be used within a certain workflow on the basis of web services. The integration of the modules into the environment of the BPMS (business process management system) solutions is possible without all know-how of the dedicated implementation.
Additionally, SIPAC organizes the scalable server farm with regard to load distribution and error handling.
SIPAC follows the SDIA principle (Software Defined Intelligence Architecture) which stands for flexibility,
modularity and scalability on standard server hardware.
The Processing Server is the central processing unit of the system. The processing capabilities can be extended using SIPAC slaves for additional hardware. The Archive Server is the data storage of the system.
The clients run the graphical user interfaces for the processing and archive services.
SIPAC can be accessed and controlled also via web services from other applications. In particular the integration into BPMS systems is possible. In fact it means that SIPAC offers automatic services for the processing and analysis of signal and data files, which alternatively could be handled by human operators.
SIPAC offers sophisticated robot services especially for the content processing and analysis of data, signals
and information from different media.

SIPAC framework
o

Platform independent

Scalable in size for required hardware configurations

User definable flow

Easy integration of user-owned and external software

SIPAC server farm

o

Parallel processing in a pool of slave-computers

Automated processing of mass data

Interactive analysis

SIPAC processing server:

o

Different analysis, processing and classification methods available

Trainable classifiers

Extendable by a user development kit (UDK)

SIPAC archive server:

o

Storage of data (e.g. inputs and processing results) and metadata

Intuitive retrieval of results

Software defined intelligence architecture (SDIA)

by MEDAV GmbH 2013 Specification subject to change Page 2/12

Impressions
Speech

Text
Automated
content
processing

Scalable
hardware
Classifier
training
support

Automated
processing

Image

SIPAC
User
specific
extensions

Interactive
analysis
Flow
management

Benefits
For the user

One interface for processing all types of information (speech, text, image etc.)

Support for automated processing

For the organisation

Higher throughput of processed data

More time for more complex tasks due to savings on routine tasks

Easy data communication / exchange possible between the departments

Amortisation of a SIPAC investment after a short time overall money saving

References
SIPAC has been installed for various customers with different requirements spanning diverse system sizes from the stand-alone workplace system based on a single high-end notebook to the
scalable system with up to 30 client workplaces on high-performance cluster architecture.
SIPAC was introduced to the market in 1999. New releases are launched every year.

by MEDAV GmbH 2013 Specification subject to change Page 3/12

The SIPAC system

SIPAC is a system for automated and interactive processing of data. The processing can be made
using different software, either provided by MEDAV, by the user or by a third party. The system is
scalable depending on the user requirements.
The situation
More and more data must be analysed by the user. Experts are busy with routine tasks. In addition, they cannot improve existing solutions. Basic users cannot apply the solutions designed by
the experts.
SIPAC is the instrument for automated and interactive processing of data. The expert may define
new flows interactively. Basic users may invoke the automated processing of mass data defined by
the experts.
SIPAC production and results can also be controlled from other software using web services.
SIPAC philosophy
The SIPAC system is very flexible and
can be configured according to the
needs of the users. The main components are shown in the following picture:
1.

In the SIPAC Processing Server,

automated processing is performed
using different classifiers, analyzers
and procedures. MEDAV provides a
variety of processing elements such
as classifiers, analyzers and procedures for content processing of
speech, text and image.

Speech

Text
Automated
content
processing

Scalable
hardware
Classifier
training
support

Automated
processing

SIPAC
Interactive
analysis

Image

User
specific
extensions

2.

Scalable hardware provides flexible

Flow
computational power involving difmanagement
ferent computers (even with different operating systems). According
to the task and the current situation,
computers can be added to or removed from the system. In the SIPAC Archive Server, all relevant data (e.g. input data, results and even the processing elements) and metadata can be
stored.

3.

Also user-owned software can be used inside SIPAC. This software can be embedded by the
user himself or by MEDAV.

4.

Flow management typically tries to compose procedures and processing elements. The goal is
to automatically analyze, filter, dispatch, channelize and process masses of data. Hereby, the
challenge is to not only plainly concatenate the stages of processing but to set the processing
into a context and relationships. This works up to complex tasks and achieves high throughputs, efficient information mining and useful limitation of the output to concentrate important information. For example, if the language of a file is determined, the following software may use
this knowledge for language specific processing.

5.

SIPAC can be used for interactive analysis, automated processing and classifier training by
calling the provided services. Services are for example file content analysis or language detection of speech samples.

6.

Trainings are performed to create or improve classifiers. A training environment is provided to

support the handling of trainable software and classifiers.

by MEDAV GmbH 2013 Specification subject to change Page 4/12

Working with SIPAC

Each data asset processed by SIPAC is handled as two inseparable files: a body file as binary data and an attribute file containing metadata.
When sending input data to SIPAC, some attributes (e.g. date, time and source) are created automatically and handled in the metadata file. During processing, the results of analyses or classifications are added to the metadata. Other important information that may come from the flow management can also be carried as metadata through the whole processing chain and is available afterwards for further use.

Following pages describe

the parts of a SIPAC system, especially the archive server, the processing server and the
clients

the software that can be embedded into SIPAC,

the configuration of the SIPAC system, and

the technical data.

by MEDAV GmbH 2013 Specification subject to change Page 5/12

System parts
A SIPAC system consists of the following software parts:

The SIPAC processing server for data processing, analysis and classification

The SIPAC archive server for storing and retrieving data

SIPAC slave(s) for distributed processing tasks to different computers and/or operating
systems

SIPAC client(s) interface for users to control the servers and tasks

Users development kit for implementing own extensions into SIPAC

For one SIPAC system one archive server, one processing server and one client is necessary.
Bigger systems may consist of for example one archive server, one processing server, 30 slaves
and 10 clients.
SIPAC processing server

The SIPAC processing server is the central processing unit of the system. More hardware can be
added for faster processing or extended processing capabilities. For each computer attached to
SIPAC, a slave software license is required. If SIPAC slaves are available with the required functionalities, the SIPAC processing server is able to distribute tasks across the attached computers
for optimal performance.
The flow management of the SIPAC processing server is based on user definable program
flowcharts (simply also known as flows) written in a XML-based command meta-language. The
command set covers all functions needed for controlling the automatic processing. The interface
for changing or creating flows is very intuitive and the system can support the user when manipulating the flows.
The SIPAC server provides an Application Programming Interface (API) with connection via
TCP/IP.
SIPAC archive server

The SIPAC archive server is the data storage of the system. All input data and metadata can be
stored there. For fast data access, it provides an up-to-date instrument of retrieval, such as categorisation of data and full-text-search.
The core functions of the archive server are:

Labelling the documents with an internal unique ID for reference

Saving and compressing files as archive documents

Storing metadata for documents (also known as attributes)

Protection of documents against changes

Configurable permission for the deletion of documents ensure archive integrity

User and password management

Access documentation by log files

Capacity and access performance is limited only by hardware

Application Programming Interface (API) connection via TCP/IP

by MEDAV GmbH 2013 Specification subject to change Page 6/12

SIPAC clients

The clients run the graphical user interfaces for both servers, the SIPAC processing server and the
SIPAC archive server. The clients are customized and provide specific graphical interfaces for the
different tasks (classifier training, flow definition, service application, monitoring, results and data
retrieval). Functions for system administration are also available.
For each operator workplace, a client license is necessary. For additional tasks, additional licenses
are necessary:

administrator/expert interface for system setup and control

training interface for the training of the involved software

All components communicate via TCP/IP protocol. If distributed machines are in use, LAN connection is required accordingly.
Tool sets

Different tool sets can be plugged into SIPAC. The tool sets can be bought from MEDAV as commercially available packages (available tool sets: Common, Speech, Text, ARGUS, Network see
next pages) or created and compiled on demand either by MEDAV or by the user himself. Third
party products can be part of a tool set. Available tool sets are described on the next pages.
User development kit

Optionally, a user development kit (UDK) is available. Using the UDK, the expert can embed own
software into the SIPAC environment. After including the customized software, the new features
will fit seamlessly into SIPAC and can be used for processing.
The UDK works command line-oriented. Programming knowledge is necessary. An expert training
is available.
Hardware

The use of hardware is very flexible: notebooks or standard computers can be used for small systems and rack-type servers for large systems. Different operating systems are supported.

by MEDAV GmbH 2013 Specification subject to change Page 7/12

Processing algorithms
A large variety of processing algorithms / software can be embedded into SIPAC.
Often it is recommended or even necessary to train the parameters of the respective software on
data of the application in order to obtain optimal performance. For audio signals for example, the
channel characteristics or language models may be trained separately.
Software can be provided by

MEDAV: offers tool sets for classification of audio, text and image processing
(others on request),
The customer himself: e.g. task specific software developed by the user,
Third party: software with a special focus, e.g. for automatic translation
To start the automated processing, files that have been sent to SIPAC are placed in a file directory
(this is a common way of production). From the file directory the files are subsequently (and/or
parallel) processed according to the user requirements. The result of processing is added to the
meta-information of the respective file.
Interactive processing provides more breakpoints for the operator to deal with inputs, procedures
and results all the way long from the original data in question to the completely processed result. In
most cases, procedures are developed, tested and evaluated in an interactive way before they are
finished to an accomplished production flow.

Tool sets
Tool sets are used as extension modules of the SIPAC system. Depending on the requirements of
the users, different tool sets can be added. Experts can create own tool sets using the User Development Kit (UDK).
Available tool sets are described in the following:
Tool set - Common

This tool set is included in all SIPAC systems. It provides the following functions:

File type detection

o More than 1.100 file types can be classified
o Depending on the file type, further processing can be made.
Packing and unpacking support, e.g. for zip files.
Basic classification: Case Based Reasoning (CBR)

Tool set - Text

The text tool set is used for the classification of files containing text (it partly includes third-party
products)

Language identification
Entity recognition
Topic spotting
Word spotting
Term translation and
Full text translation
Optical Character Recognition (OCR)

Some of the algorithms are language specific and, therefore, training data must be collected
beforehand.

by MEDAV GmbH 2013 Specification subject to change Page 8/12

Tool set - Speech

The speech tool set is used for the classification of audio files. Referring to different tasks, the
audio signal is analysed / processed with respect to the following criteria:

Speech detection
Gender identification
Language identification
Speaker identification
Topic spotting
Word spotting
Speech-to-phoneme
Speech-to-text

Some of the algorithms are language specific and, therefore, training data must be collected
beforehand to improve the classifiers for optimal performance.
Tool set ARGUS

ARGUS is a special analysis tool set for images. It provides the following functions:

Support Vector Machines (SVM) as a general binary classification method

Concept of generic content based classification (collecting positive and negative samples,
defining feature extraction, automated SVM-training and -test, SVM-classification)
Image file format specific analyzers (check for file manipulations; for jpeg, png, bmp, others
upon request)
Trained classifiers for bitmap- and jpeg-steganography (StegoBitmap and StegoJPG; check
for hidden information in images)
Trained classifier for high textual content (HiTex) in images (independent of the lettering
system)
Heuristic malware detection (retrieval for indication of potential malicious code without
using signatures)
Optical Character Recognition (OCR)

Tool set STRIX

STRIX is a special analysis tool set for captured network data streams. It provides the following
functions:

Detection of encrypted traffic in binary network streams

Support of Deep Package Inspection (DPI)

by MEDAV GmbH 2013 Specification subject to change Page 9/12

Working with SIPAC

Following, an example of how to work with SIPAC using the tool set speech is given.

Employ the needed software

Define the order of processing

The configuration is done in the following steps both MEDAV and the users are able to configure
the system.
Task analysis
The task has to be defined exactly with respect to classifiers and data volume.
Example: A system shall be configured for the classification of the prevailing language in audio
files. The signals are obtained from the telephone line.

Languages of interest: English, German, Spanish, French, Italian, Danish

Expected data volume: 1000 hours of audio signals to be produced per day.

This leads to a need of speech classifiers for the requested languages and a calculation of the
necessary amount of processing hardware.
Choice of available classifiers
If trained classifiers are available for the given task, nothing has to be done. Otherwise, training
data must be collected (or purchased) and the classifiers must be trained.
Example: Trained classifiers are available in telephone quality for all languages but Danish. Thus,
Danish telephone data must be collected. A classifier for Danish must be trained.
Configuration of the flow
The workflow describing the sequence and conditions of the automated processing must be established.
Example: After speech detection, language identification is performed for the parts of the audio
signal containing speech.
Test and evaluation
The flow is activated including the embedded classifiers. Testing is performed in order to ensure
correct function and high quality.
Example: Run SIPAC with a set of test files. The evaluation of the language classification results is
performed.
After these steps, the SIPAC system is ready for use and can be started in the user scenario.
For optimal performance, it is recommended to perform the evaluation on a yearly basis in order to
maintain the high accuracy with respect to changes of the incoming data (e.g. different telephone
codecs).

by MEDAV GmbH 2013 Specification subject to change Page 10/12

Technical data
General

The framework is scalable from a standalone solution to a server farm

Internal communication via TCP/IP

Supported operating systems

Windows: XP, Server 2003, Vista, Server 2008, 7

Linux

Solaris

Others upon request (depending on Java platform)

SIPAC Processing Server

Parallel processing of a not unlimited but sufficient number of processes

Communication with the SIPAC Archive Server (read, write)

Local storage of intermediate results

Integration platform and framework for external tools

Slave:
o

distributed processing according to slave configuration / load

implemented tools can run on different operating systems

SIPAC Archive Server

Storage of data and assigned metadata

Storage of flows

Storage of classification and intermediate results

Storage of processing elements such as classifiers, analyzers and procedures

Search functionality

SIPAC client

Client application (Fat Client)

Graphical user interface for system control

Simple role concept: administrator, user

User Development Kit

Apache Ant based tool for compiling tool sets

Command-line-oriented

by MEDAV GmbH 2013 Specification subject to change Page 11/12

Corporate Policy

Sensors

Technology
in the products, development and in the company management
is state-of-the-art and represents a top level.
Quality
... in all divisions of our company is considered as the indispensable
prerequisite for a risk-free and successful cooperation with our customers
and business partners.

Signals

Position in the market

is affected by extensive experience gained from signal and information
processing. We are prepared best to face competition.

Classification

Product and engineering spectrum

are comprehensive, complete and tailored to meet all requirements.
As a single source supplier of solutions, we offer standard devices,
systems and services.
Employees

Content

form the roots of the company and render the services necessary for
maintaining and expanding the technical basis and a trustful and fair
cooperation.
Growth

Information

... on a stable technical and economical basis at home and abroad is

our declared long-term goal.
Trust and fairness
vis--vis our business partners and within the company are the basis
of our business.

Intelligence

Compliance
... with excessive sensibility and compliance with German and international
export regulations we act on a worldwide basis.

MEDAV GmbH Grfenberger Str. 32-34, D-91080 Uttenreuth Homburger Platz 3, D-98693 Ilmenau
Phone +49 9131 583-0 Fax: +49 9131 583-11 E-Mail: info@medav.de www.medav.de

by MEDAV GmbH 2013 Specification subject to change Page 12/12

w713od.065