Skip to main content

Speech Interface at Office Workstation

Objective

The overall aims of the SPIN project were to:
-make significant advances in speech input/output algorithms
-study ergonomic aspects in relation to the integration of speech input/output facilities in the office environment
-build a demonstrator illustrating the main results of the project, consisting of a prototype of a workstation with speech facilities.
The aims of the project were to make significant advances in speech input/output algorithms, study ergonomic aspects in relation to the integration of speech input/output facilities and build a demonstrator.

The MPLPC algorithm was the coding method chosen and a simulation of refined versions and a real time breadboard implementation of the algorithm are available.
Work concentrated on algorithm studies. The results were embodied in an experimental demonstrator, MARIPA, which was a low cost recognizer based on a personal computer (PC) board. The demonstration of the first stage of the continuous speech recognizer, producing a lattice of demisyllables for continuous speech recognition, was achieved. Speech input/output assessment methodology was refined and used to improve the quality of the developed algorithm.

Text to speech synthesis systems for French, Italian and Greek were developed, with emphasis on phonetic components, prosody, development of a speech rule compiler and quality evaluations of the French and Italian diphone sets.

Automatic test software for the simulation of speaker verification was produced, running on a large speech database and a final real time automatic speaker verification system was implemented.

A 3 digital signal processing (DSP) based modular hardware for speech processing was designed and developed, and each speech algorithm implemented. An experimental final system, made up a PC and of a speech interface was built to present the results of the project.
Experiments were completed and results and guidelines delivered with regard to:
integration of speech input/output facilities in a multimedia person machine interface;
use of speech in specific office applications;
definition of the best positions for a microphone on a workstation for speech input;
definition of a measuring technique for determining the noise sensitivity of speech recognizers;
behaviour of test subjects when using a multimedia user interface.
The main results of the SPIN project were as follows:
Speech Algorithms
-Coding
Three coding methods (MPLPC, TDHS, and RELP) have been studied in detail and simulated versions of these algorithms produced. Evaluation of their intelligibility showed that the MPLPC algorithm was the best one at the chosen bit rate (9.6 Kbit/s); a simu lation of refined versions and a real-time breadboard implementation of the MPLPC algorithm are currently available.
-Speech recognition
Work was concentrated on algorithm studies. The results were embodied in an experimental demonstrator, MARIPA, which was a low-cost recogniser based on a PC board. The demonstration of the first stage of the continuous speech recogniser, producing a latt ice of demi-syllables for continuous speech recognition, was also achieved. Speech input/output assessment methodology was refined and used to improve the quality of the developed algorithm.
-Text-to-speech synthesis
Text-to-speech synthesis systems for French, Italian and Greek were developed, with emphasis on:
.phonetic components: full diphone dictionaries are available for the three languages dealt with in the project
.prosody: many rules of duration and intonation were defined
.development of a speech rule compiler
.quality evaluations of the French and Italian diphone sets.
-Speaker verification
Automatic test software for the simulation of speaker verification was produced, running on a large speech database. A final real-time automatic speaker verification system was implemented and used to control access to protected areas of the R&D laborato ries.
Hardware Implementation and Integration
A three DSP-based modular hardware for speech processing was designed and developed, and each speech algorithm (coding, speech recognition, speaker verification, text-to-speech synthesis) implemented.
An experimental final system, made up a PC and of a speech interface (built in a VME environment and connected to the PC via a serial line) was built to present the results of the project. The office application chosen was agenda planning.
Ergonomic aspects were also carefully studied. Several experiments were completed and significant results and guidelines delivered with regard to:
-integration of speech input/output facilities in a multimedia person-machine interface
-use of speech in specific office applications
-definition of the best positions for a microphone on a workstation for speech input
-definition of a measuring technique for determining the noise sensitivity of speech recognisers
-behaviour of test subjects when using a multimedia user interface.
Exploitation
The results of this project were used in project 954, IKAROS.

Coordinator

SOCIETE ETUDES SYSTEMS AUTOMATIONS (SESA)
Address
3 Rue Du Clos Courtel
35018 Rennes
France

Participants (9)

Alcatel Alsthom Recherche
France
Address
Route De Nozay
91460 Marcoussis
CMSU-COMMUNICATION & MANAGEMENT SYSTEMS UNIT.
Greece
Address
Zographou Campus
15773 Athinai
COMMISSARIAT A L'ENERGIE ATOMIQUE
France
Address
Rue De La Federation 31-33
75015 Paris
Centro Studi e Laboratori Telecomunicazioni SpA
Italy
Address
Via G. Reiss Romoli, 274
10148 Torino
Daimler-Benz AG
Germany
Address
Wilhelm-runge-straße 11
89013 Ulm
Oros SA
France
Address
13 Chemin Des Prés Zirst
38241 Meylan
SIEMENS-NIXDORF INFORMATIONSSYSTEME AG
Germany
Address
Berliner Straße
1000 Berlin
Scuola Normale Superiore di Pisa
Italy
Address
Piazza Dei Cavalieri 7
56126 Pisa
UNIV VAN AMSTERDAM
Netherlands
Address
Roetersstraat
1018 Amsterdam