RESPITE: Recognition of speech by partial information techniques

• Research unit: EU RFP
• Project number: 98.0086

Key words (English)

Speech recognition; hidden markov models; missing data theory; subband based speech recognition

Alternative project number (English)

EU project number: 28149

Research programs (English)

EU-programme: 4. Frame Research Programme - 1.3 Telematic systems

Short description (English)

See abstract

Further information (English)

Full name of research-institution/enterprise:
Institut dalle Molle d'intelligence artificielle perceptive IDIAP
Research Institute

Partners and International Organizations (English)

Sheffield University (UK), Daimler-Chrysler (D), Faculté Polytechnique de Mons (B), ICP (F), Babel technology (B)

Abstract (English)

The RESPITE project aimed at developing techniques for automatic speech recognition that are truly robust to unanticipated noise and corruption, with possible car applications (with Daimler-Chrysler as our main industrial partner). The techniques developed and tested in several conditions were based on a combination of emergent theories of decision-making from multiple, incomplete evidence sources and of human speech perception. More specifically, new recognition paradigms based on multi-stream processing and the missing data theory have been investigated in much detail and tested on different databases (including in the framework of international evaluation campaigns), on real data (recorded in car environments), as well as part of a fully integrated in-car system.
At the end of the RESPITE project, the main conclusions are that the resulting system based on missing data and multi-stream speech recognition techniques are more robust to noise, without requiring specific noise adaptation or noise model. Being adopted by numerous labs, it is also expected that these new techniques could yield advances in adjacent recent fields, such as the handling of multiple temporal resolutions and the processing of multi-modal information (e.g., audio-visual fusion

References in databases (English)

Swiss Database: Euro-DB of the
State Secretariat for Education and Research
Hallwylstrasse 4
CH-3003 Berne, Switzerland
Tel. [phone]
Swiss Project-Number: 98.0086