Perspectives on Speech Separation — a Workshop
October 31 - November 2, 2003
Montreal, Quebec
sponsored by the National Science Foundation

[objectives] [topics]


An overlook on the various methods makes the observer conclude that at least some of the limitations of each of the schemes could be overcome by integrating them into a single approach. Such integration, however, has the prerequisite that proponents of different approaches get together, present their methods, theories, and data, discuss these openly, and attempt to find ways to combine schemes, in order to achieve the ultimate goal of separating speech signals by computational means. Given the sheer volume of current work on the diverse methods, such a meeting of researchers is not only necessary but also extremely timely. The proposed workshop, to our knowledge first of its kind, intends to be an undertaking to reach this objective. Another benefit of this workshop is that investigators of different fields, or adepts of different approaches, may decide to explore possibilities of attacking the problem jointly. While the development of such collaborative ventures cannot be planned, the workshop may represent an opportunity for instances of such collaboration to be born – collaboration between computer scientists working in different directions on the problem as well as collaboration between experimental psychologists/speech scientists and computer scientists.

The primary impact of the workshop thus lies in its attempt to treat the difficult and timely problem of speech separation in an interdisciplinary fashion, recognizing that solution of the problem may need an approach that consists of threads of knowledge, models, and proposals from different scientific/technological fields and areas. Since the workshop is multidisciplinary in nature, it necessarily will have an educational impact by exposing participants to results, theories, and methods from fields that are not their own. The workshop is also expected to serve as an opportunity for the participants to develop professional ties possibly leading to collaborative research plans.

The broad impact of the workshop lies in its attempt to build two communities. The first community is that of scientists dedicated to working toward solutions of speech separation within their own field, hoping that the dialog commenced at the workshop will live on. Because the participants are also engaged in teaching and training students, postdoctoral fellows, and junior investigators, the workshop is also bound to have the impact of passing on to the younger generation of scientists the importance of appreciating the need for, as well as the difficulty of the problems of speech separation. Building this second community will take time but the workshop is expected to contribute to the effort.


The workshop will address speech separation as viewed from different standpoints: experimental psychology, neurophysiology, computational modeling of behavioral and neurophysiological data, and diverse computational directions designed to achieve automatic separation of speech sources. Accordingly, the specific topics discussed will include

• behavioral experiments on the auditory scene analysis of multiple speech sources;
• experiments on human central nervous system activity while performing speech separation;
• theories and computational models of listeners’ performance as regards diverse aspects of the “cocktail-party phenomenon”;
• bottom-up and top-down processes in speech separation;
• computational auditory scene analysis;
• learning and speech separation;
• the role of spatial separation of speech sources;
• statistics-driven and other blind-source separation techniques applied to speech separation;
• separation of acoustic signals compared to the separation of graphic objects.

Presentations grouped by topics will be followed by moderated discussion periods giving opportunity to all participants to engage in an exchange of views. A comprehensive discussion period at the end of the workshop will permit the participants to discuss the general current status as well future directions of the field of speech separation.


[top of page]