|
PlainSpeech
Speech Recognition
For Dictation Broadcast and Telephony
Speech is the most natural form of communication. Profit from years of
experience from
researchers at the Aachen Institute of Technology who have worked on the
interface
between man and machine! Our modules form the basis for individually
tailored speech
recognition systems that are easily integrated into your products.
PlainSpeech offers a new technology that significantly improves and
simplifies the use
of technical products:
• Fast and reliable recognition of continuous speech
• User specified speaker-dependent or speaker-independent recognition
• Gender-independent speech recognition
• Scalable vocabulary, scalable number of recognized languages
• Grammar and speech models to improve speed and quality
• Dialogue guiding with dynamic database interface
The system manages almost any recognition
task: from simple number chain
recognition and complete call center solutions to information retrieval with
vocabularies
of 100,000 words or more.
Command word
recognition
High speed and remarkably low error rates characterize our speech
recognition for
small vocabularies. The system can easily be expanded any time with new
commands
to adapt to changing needs and demands.
Applications for this area are for example:
• Speech commissioning in logistics
• Checking and recording in quality assurance work
• Guidance and navigation interface for user terminals
• Speech driven consumer applications
Dialogue System / ''Interactive Voice Recognition (IVR)''
Dialogue systems are scalable.
Command word recognition modules can simulate dialogue processes well in a
decision tree structure. PlainSpeech provides sufficient freedom in the
definition of dialogue processes and provides reliable recognition even in
noisy environments. Lengthy dialogues that might occur in a call center can
also be accommodated. Standard questions can be answered by automatic
systems at any time of day. Qualified employees can concentrate on their
primary tasks in manning customer relations. PlainSpeech efficiently
processes customer dialogues and makes it easy for customers to navigate
telephone information and ordering
systems.
Recognition of continuous speech
Thanks to its scalability AppTek's speech recognition engine can also be
expanded into a dictation system with a large vocabulary. Especially good
results are achieved for tailored solutions for specific sectors.
Performance and quality are assured through the
use of the newest speech and grammar models. Further features include
voice-to-email or voice-to-SMS. One application for the recognition of
continuous speech is the Multilingual Information Retrieval: In light of the
many languages on the World
Wide Web, a language-independent system can help avoid being limited by
one’s
mother tongue. The development of such a self-learning system is another
task of AppTek.
Multilingual Information Retrieval makes it possible to ask questions
pertaining to a
particular topic, find relevant documents in other languages, summarize
these and translate them into the desired language. All relevant
information becomes accessible
and comprehensible for the user. The system is initially being developed for
the eleven
relevant EU languages, and will later be expanded to other languages.
A further application is Spoken Document Retrieval: The basis for this
application is
the combination of speech recognition PlainSpeech with modules for text
classification
and extraction from the Knowledge Management area of AppTek.
The system improves the use of audio mediated databases. For example,
Internet users
can find spoken news. Media companies can do automated searches of their
archives.
Searches can be entered in text in the usual manner. A combination of the
speech
recognizer PlainSpeech with the modules from AppTek’s machine translation
area
creates a speech-to-speech translation system.
This system, that received awards as part of the VERBMOBIL project, enables
automatic
live translation for specific content domains. Our collection of languages
and supported
content domains grows continuously.
Technical specifications:
The PlainSpeech-system is a current generation self-learning
phoneme/grapheme-speech-recognizer.
Supported platforms (Operating system, architecture):
Microsoft Windows NT, 2000, XP, Sun Sparc Solaris, Linux, StrongARM, XScale,
MacOSx, Power PC
Supported software interfaces / Standard-APIs: C/C++, Java, XML / SGML
Supported standards: VoiceXML, Salt, W3C-XML-Standards for grammar and
lexicons,
Speech Assessment Methods Phonetic Alphabet (SAMPA)
We offer the scalable possibility of using the following grammars to develop
language
models:
Statistical grammar for commando word recognition, Finite State Automaton (FSA),
Standard n-gram and gap-n-gram language models, combinations of these
language
models, context free grammar models (CFG).
SpeechWay Platform
PlainSpeech.WAY is software that integrates telephone, computer and other
communications modes into one modern system. Through integrated speaker
independent speech recognition, the system receives all commands through
speech
input.
 Telephone switchboard –
tailored to customer needs PlainSpeech.WAY can be used as an intelligent
phone switchboard
that routes calls, takes messages, delivers information, interoperates with
databases, sends faxes, and much more. Each of these functions is
represented visually on the graphical user interface. The representation is
very user friendly and has a very steep learning curve. The user can set up
the various functions according to his own individualized needs.
The different functions are connected to each other and can be used as often
as necessary. A script can be developed for any scenario, and these are
easily edited and expanded.
Telephone switchboard – fully automated
Callers can now use speech commands to navigate such a script.
PlainSpeech.WAY recognizes spoken words and activates the desired function.
For example, after saying "Support" the caller is connected to the tech
support department of the company, or by saying "record a message" he can
leave a voicemail for an employee. The capabilities of speech recognition
also permit customers who don’t have touch-tone capabilities, or don’t want
to use these, to take advantage of modern telecommunications functions. The
ability to dial out by
usual methods remains – any function can be driven by speech commands or
touch-tone.
PlainSpeech.WAY can also be used for intelligent call holding. While the
user is speaking on another line, the person waiting can receive other
information.
The PlainSpeech.WAY user is reachable anytime:
and if necessary can be reliably represented by PlainSpeech.WAY. Telephone
switchboard – that meets any demand. The target person of a voicemail
can be informed by mobile phone of a message waiting immediately or at a
time specified by the recipient. Messages can also be forwarded by email
with a Wave attachment. Using caller-ID, messages can be routed in different
ways through personalized functions. This way reaching the appropriate
function is made more efficient.
An additional function allows users to launch other applications on a
computer through a script. For example, if an office security system is
controlled through a computer, this can be turned on and off through a phone
call.
Functions that are not supposed to be freely accessible can be protected
through spoken
passwords or touch-tone PINs. Calls are logged so that requests can be
traced anytime. Dialogue guiding with dynamic database interface. The
application of PlainSpeech.WAY is by no means limited to a single script.
Depending on time of day different scripts can be used to
guide the dialogue with the caller and the PlainSpeech.WAY software. The
simplest example is to greet the caller with "Good morning" ,"Good
afternoon" or "Good evening“depending on what time it is.
Telephone switchboard – easily implemented
PlainSpeech.WAY can be used with any standard PC running Windows. The
connection between computer and phone line is made through an ISDN
interface. All
additional functions are provided by the PlainSpeech.WAY software.
Almost all activity is run in the background, such that PlainSpeech.WAY does
not use
up valuable computing resources. The computer can be used for other
activities.
Telephone switchboard – usable in any company
 The
friendly voice from PlainSpeech.WAY greets the caller who can determine the
course of the call through voice commands. Calls are easily connected
without robbing
the caller of valuable time waiting to be put through. If the caller only
wants to obtain
information about services offered (e.g., current theater schedule) or take
advantage of
a specific offer (e.g., order theater tickets), he can easily accomplish
this through
automated transfer of the call to the appropriate person. PlainSpeech.WAY
combines
the services of a professional answering service with the functions of an
answering
machine, while also having the advantage of offering different user
determined voice
messages.
PlainSpeech.WAY – brief overview
PlainSpeech.WAY can accept up to 16 calls at the same time. The functions
can provide the following advantages:
speaker independent speech recognition
Database interface
Text-To-Speech
User defined dialogues
Individualized and standard messages
User friendly interface
Interface with external applications
uses standard PC Hardware
ISDN technology
Call forwarding
Caller identification by caller number
Automatic dialing
Time controlled dialogs
Variable handling
The functionality of PlainSpeech.WAY enables working closely with customers
and increases the competitiveness of a company by:
optimizing customer interactions
targeting phone conversations
intelligent phone dialogues
flexible call holding
automated routine tasks
unlimited reachability
|
|
