REALISM 2018 - Special Issue on REALISM IN ROBUST SPEECH AND LANGUAGE PROCESSING

How can you be sure that your research has actual impact in real-world applications? This is one of the major challenges currently faced in many areas of speech processing, with the migration of laboratory solutions to real-world applications, which is what we address by the term "Realism". Real application scenarios involve several acoustic, speaker and language variabilities which challenge the robustness of systems. As early evaluations in practical targeted scenarios are hardly feasible, many developments are actually based on simulated data, which leaves concerns for the viability of these solutions in real-world environments.
Information about which conditions are required for a dataset to be realistic and experimental evidence about which ones are actually important for the evaluation of a certain task is sparsely found in the literature. Motivated by the growing importance of robustness in commercial speech and language processing applications, this Special Issue aims to provide a venue for research advancements, recommendations for best practices, and tutorial-like papers about realism in robust speech and language processing.
Prospective authors are invited to submit original papers in areas related to the problem of realism in robust speech and language processing, including: speech enhancement, automatic speech, speaker and language recognition, language modeling, speech synthesis and perception, affective speech processing, paralinguistics, etc. Contributions may include, but are not limited to:
- Position papers from researchers or practitioners for best practice recommendations and advice regarding different kinds of real and simulated setups for a given task
- Objective experimental characterization of real scenarios in terms of acoustic conditions (reverberation, noise, sensor variability, source/sensor movement, environment change, etc)
- Objective experimental characterization of real scenarios in terms of speech characteristics (spontaneous speech, number of speakers, vocal effort, effect of age, non-neutral speech, etc)
- Objective experimental characterization of real scenarios in terms of language variability
- Real data collection protocols
- Data simulation algorithms
- New datasets suitable for research on robust speech processing
- Performance comparison on real vs. simulated datasets for a given task and a range of methods
- Analysis of advantages vs. weaknesses of simulated and/or real data, and techniques for addressing these weaknesses
Papers written by practitioners and industry researchers are especially welcomed. If there is any doubt about the suitability of your paper for this special issue, please contact us before submission.
*Submission instructions: *
Manuscript submissions shall be made through EVISE at https://www.evise.com/profile/#/SPECOM/login
Select article type "SI:Realism Speech Processing"
*Important dates: *
March 1, 2017: Submission portal open
May 31, 2017: Paper submission
September 30, 2017: First review
November 30, 2017: Revised submission
April 30, 2018: Completion of revision process
*Guest Editors: *
Dayana Ribas, CENATAV, Cuba
Emmanuel Vincent, Inria, France
John Hansen, UTDallas, USA