Designing New Speech Interfaces with Ambient Audio

Funding source

Swedish Research Council - Vetenskapsrådet (VR)

Project Details

Start date: 01/01/2017
End date: 31/12/2020
Funding: 3400000 SEK


While speech recognition and audio feature extraction have made
significant steps forward in recent years, in particular through the
deployment of new recognition and modelling techniques, human computer
interaction (HCI) issues around speech have been slower to develop.
Moreover, commercial speech responsive agents, such as Apple’s Siri, and
Microsoft’s Cortana, are almost entirely focused on the use of
dialog-based models of interaction with users with restricted sequences
of interaction. Yet microphone equipped devices such as mobile phones
are continually surrounded by spontaneous person-to-person talk and
ambient environment audio. In this project we aim to harness this
ambient audio, first exploring what can be computationally detected from
such an audio stream then proceeding through a ‘research through
design’ process to build, deploy, and test new systems and services
based on always-listening devices. We will draw upon advances in
conversation analysis developing new recognition techniques focused on
noisy large datasets of ambient audio. Outcomes of the project will
include four demonstrator systems, as well as reengaging the HCI
community with new ways of designing systems for speech.

Last updated on 2017-05-10 at 07:45