Intelligence Environment
Speech Detection
Language Identification
Speaker Identification

– Subject to–change
notice!prior notice!
to change


MELANIE - Intelligence Environment

Why Speech Classification?
The increasing quantity of speech
signals needs an effective and high
quality means for the analysis of
audio data.
Obtaining support from an automated software tool is highly
appreciated by operators working
with the analysis of speech files.
MELANIE can help support the
operator with:
• Automated analysis of incoming
speech signals
• A means for analysis of incoming
speech signals using the results of
the MELANIE classifiers.

What Can Be Analysed?
MELANIE provides a software library for the classification of incoming speech
signals referring to
• Speech Detection:
The parts of audio signals containing speech are marked. For storing the incoming
signals, only those parts containing speech may be stored.

• Language Identification:
All incoming audio signals are classified according to the speech prevailing in the
audio files. All languages of the world can be identified as soon as training material
is available.

CONFIDENTIAL – Subject to change without prior notice!


• Speaker Identification:
Speakers in incoming audio signals
can be identified. The algorithm is
independent of the spoken language,
thus a speaker can be found also
when communicating in another
Training material for the speaker to be
analysed is necessary, a rejection of
unknown speakers is made.

Product Description
For these classifiers, training, testing, and automatic production can be performed.
For each application, libraries of software functions are available, together with a
standardized and open interface (CORBA).
• Training environment: Classifiers can be trained optimally for the domain of the
user. In a simple script-based training, the user can build the classifiers according
to his domain and data material.
• ELAMAN offers trained models for some applications, the user does not have to
train the classifier’s parameters.
• Production environment: This environment automatically classifies all incoming
audio signals using the available classifiers. A simple standardized interface is
given for the automatic production 24/24h. The production environment is available
as PC license or as site license.
The classifiers can be used via the CORBA interface. They can be integrated into
PC-based OEM applications.
Classifiers are available for the following applications:
SQ Speech
MP Language
GMM Speaker

Detection and Language Identification

Technical Data
Input format

Real signal comprising
• Sample Rate:
• Data Type:

8, 16 kHz
PCM, A-law, μ-law

Also available as options, routines for processing other data

CONFIDENTIAL – Subject to change without prior notice!



The classification result can be accessed from the output
interface. The output is provided as:
• Classification results in equidistant configurable time
segments, e.g. for generating a label track (maximum score,
individual score for all classes)
• Classification results for the entire signal up to the signal end
(maximum score, individual score for all classes)


Standard interface CORBA enables the access to all results of


The classifiers show reliable classification accuracy combined
with a high processing speed. Realtime factor: Depending on
the type of classifier and the amount of classes (such as
languages, speakers), ranging from 1/7 to 1/20 with e.g.:
• Hardware: P4; 3 GHz; 1GB RAM
• Different Classes: 10

Company Principles and Policy
… in development and company management is state-of-the-art,
and represents only the best.
… in all areas of our company is regarded as the almost requirement
for risk-free and successful cooperation with our customers,
and business partners.
Market Position
… we are the specialists in the field of signal and data processing
as well as pattern recognition, and we are glad to face competition.
… form the roots of our company, and give the performance required
for maintaining and building the technical base, and close
personal cooperation we have with our clientele.
… we strive toward a healthy, stable foundation at home and
… are comprehensive and complete. As a full-system company
we offer standard equipment, systems, and services.
… in the relationships to our business partners, and within our
own company forms the basis of our business.

CONFIDENTIAL – Subject to change without prior notice!


Consequently, this means that a recording with a signal length of 20
minutes can be fully classified in one minute.

CONFIDENTIAL – Subject to change without prior notice!


Document Path: ["melanie-intelligence-environment.pdf"]


Click to send permalink to address bar, or right-click to copy permalink.

Un-highlight all Un-highlight selectionu Highlight selectionh