| The Fully Networked Car, A Workshop on ICT in Vehicles |
A B S T R A C T |
|
| Speaker: |
Sebastian Möller
Institute of Communication Acoustics (IKA), Ruhr-University Bochum, Germany |
| Session 7-I: |
In-Vehicle devices and Human Machine Interface
(including Voice Interface) |
| Session Moderator: |
Jean-Yves Monfort, Chairman of ITU-T Study Group 12, France Telecom & Jean-Pierre Cheynet, Chairman of ISO TC-22 "Road vehicles", Bureau de Normalisation de l'Automobile |
| Title of Presentation: |
Evaluation of In-Car Voice Services: Tasks for the New Q.12/12 |
|
Speech technology, including speech and speaker recognition, spoken language understanding, dialogue management and speech synthesis, has become a major enabling technology for communication in the car environment. The services which are based on speech technology range from simple voice-dialling over car navigation to more complex office applications (e.g. email reading) and control of remote domestic devices. The car environment puts considerable demands on the performance on the underlying speech technology modules, namely in terms of robust speech recognition, intelligible speech synthesis, and an interaction which only minimally distracts from driving.
The ITU-T recently addressed these problems by setting up a new question on the “Performance Evaluation of Services Based on Speech Technology” (Q.12/12). On the one hand, this question will investigate the impact of acoustically adverse environments and degraded transmission channels on speech and speaker recognition performance. First results have already been presented, and they will be compared to the predictions of models which are commonly used for network planning (i.e. the E-model) or for instrumental quality prediction (e.g. Rec. P.862). New investigations are expected for quantifying the quality of synthesised speech, e.g. with the help of models recently developed by ITU-T SG12 (Rec. P.563).
On the other hand, perceived quality and usability of entire services which are based on speech technology will be measured. A first Recommendation on subjective evaluation methods has recently been approved (Rec. P.851), and it is foreseen that additional guidance is needed for usability evaluation of such services, in particular for in-car applications. In parallel, proposals have been made to quantify service performance in a parametric way. The talk will give an overview on these activities, and it will point out which aspects are of particular importance when speech technology is used in a car environment.
|
|
|
|
|
