Publikation
Automatic Speaker Age and Gender Recognition in the Car for Tailoring Dialog and Mobile Services
Michael Feld; Felix Burkhardt; Christian Müller
In: Proceedings of Interspeech (2010). Conference in the Annual Series of Interspeech Events (INTERSPEECH-2010), September 26-30, Makuhari, Japan, ISCA, 9/2010.
Zusammenfassung
Car manufacturers are faced with a new challenge. While
a new generation of "digital natives" with a natural affinity to computers
and Internet services has grown up and become a new customer
group, the problem of aging society is still increasing. This
emphasizes the need of providing flexible in-car dialog and access to
mobile services that take into account the specific needs and preferences
of the respective user (group). In order to tackle this issue, three
top-level problems have to be solved: 1. How do we find out which
group the current user belongs to?; 2. How should the knowledge be
represented and linked to knowledge on the system / service in order
to support adaptation?; 3. What would be the best adaptation strategy?
In this paper, we address the first question. We consider speech
as one of the possible sources of information that allow non-intrusive
user model acquisition. We present a GMM/SVM-supervector system
(Gaussian Mixture Model combined with Support Vector Machine)
for speaker age and gender recognition, a technique that is
adopted from state-of-the-art speaker recognition research. We furthermore
describe an experimental study with the aim to evaluate
the performance of the system as well as to explore the selection
of parameters. In comparison to previous work on age and gender
recognition (by the authors and others), the underlying task is considered
more difficult because we modified the definition of the age
classes and used particularly short utterances for testing. Nevertheless,
a comparable accuracy has been obtained. Additional contributions
of this paper are: a structured itemization of experimental
results, which shed light on the effect of various design decisions,
as well as a concrete conceptual outline with respect to problem two
(knowledge representation) and three (adaptation strategies).