Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Speech-to-Speech Translator for Natural Language

IP.com Disclosure Number: IPCOM000111307D
Original Publication Date: 1994-Feb-01
Included in the Prior Art Database: 2005-Mar-26
Document File: 4 page(s) / 169K

Publishing Venue

IBM

Related People

Sharman, R: AUTHOR

Abstract

A device is proposed which translates a speech act in one language into an equivalent speech act in another language, in real time.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 38% of the total text.

Speech-to-Speech Translator for Natural Language

      A device is proposed which translates a speech act in one
language into an equivalent speech act in another language, in real
time.

      A person in one country cannot speak to a person in another
country, unless they share a common language.  This restricts the
volume of telephone traffic and the types of services which can be
provided.  A solution is to build a prototype Speech-to-Speech
translator from custom parts, consisting of Speech Recognition,
Translation Manager, and Speech Synthesis components to result in a
sub-optimal communication tool.  An interactive architecture for a
Natural Language (NL) interlingual speech translating tool would
enable the Speech and Translation components to be used in a way
which would make robust, cooperative dialogues possible.

      The market for international telephone calls is large [5], with
increasing traffic taking place across national language boundaries.
The highest traffic loads are only achieved where countries share the
same language (e.g., US-UK, Germany-Austria) or where there are large
populations of emigrant workers (e.g., US-Mexico).  Traffic across
national boundaries where a language is not shared is much lower.
Telephone interfaces are also undergoing a revolution in the design
of dialogues [1], stratagems and other techniques for achieving
effective communication between human and automatic services [2].
Callers, are becoming more ready to accept automated, or partially
automated, conversations [3], but expect a high level of service from
telephony interfaces for a variety of services [4].  The class of
service proposed is based on existing Speech and Telephony
technologies, which will provide a real-time Speech-to-Speech
translating service for human respondents who do not share the same
Natural Language.  Current solutions for this problem are listed
below:

1.  Restriction to non-native language:  one, or both, parties in the
    conversation uses his/her non-native language and may be
    restricted by fluency in that language.

2.  Use of a bilingual agent: another, third party, person holds the
    conversation for one of the parties.  This may be expensive,
    difficult to arrange, and infringe on confidential or personal
    information.

3.  Restricted service quality: this may be provided by automated
    phone response systems, using DTMF tones or similar to
    communicate information.  This is awkward, limited and requires
    good design of the application dialogues, which does not yet
    exist.

4.  Other methods: alternatives usually involve avoiding
    voice-to-voice contact altogether, for example, by using email,
    FAX, surface mail, etc..  These methods are non-real time, slow
    and not interactive.

      The architecture of a basic system would be as shown in Fig. 1.
The functions could be implemented in the following ways: all local...