DAGA 2010 - Deutsche Gesellschaft für Akustik eV

244 DAGA 2010 Programm
Do. 15:15 Atze-Theater ANC/AVC
Robust Active Noise Control (ANC) for Engine Booming Noise and
HVAC Noise Reduction
Y. Naor, K. Kirshner und A. Bereby
Silentium, Rehovot (Israel)
The noise inside a vehicle cabin is a combination of several noise
sources: road noise, wind noise, ventilation noise and noise generated
by the engine. The engine noise is transmitted to the cabin as air-borne
and structure-borne noise; as a result, periodic noise components are
excited inside the vehicle cabin. In the scope of this work, two ANC modules
were developed and implemented. The first ANC module is an
active muffler assembled on the inlet/outlet of the HVAC unit, aiming to
reduce the ventilation noise. The second ANC module was designed to
reduce the noise excited by the engine. Both modules were realized on
the same Silentium’s S-Cube TM controller. In this paper, we discuss the
challenges of implementing ”real-world” ANC controller for the automotive
industry. The following results can be reported: HVAC unit - 15dBA
[SWL] reduction, Engine unit - 20dBA of noise reduction at the second
harmonic while driving with constant speed condition, and 10-15dBA of
noise reduction was obtained at the second harmonic while the speed
of the engine changed from 1000RPM to 4000RPM within 10 seconds.
Do. 8:30 Gauß B 501 Robuste Spracherkennung
Options for Modelling Temporal Statistical Dependencies in an
Acoustic Model for ASR
V. Leutnant und R. Haeb-Umbach
Universität Paderborn
Traditionally, ASR systems are based on hidden Markov models with
Gaussian mixtures modelling the state-conditioned feature distribution.
The inherent assumption of conditional independence, stating that a feature’s
likelihood solely depends on the current HMM state, makes the
search computationally tractable, nevertheless has also been identified
to be a major reason for the lack of robustness of such systems. Linear
dynamic models have been proposed to overcome this weakness
by employing a hidden dynamic state process underlying the observed
features. Though performance of linear dynamic models on continuous
speech/phone recognition tasks has been shown to be superior to that
of equivalent static models, this approach still cannot compete with the
established acoustic models.
In this paper we consider the combination of hidden Markov models based
on Gaussian mixture densities (GMM-HMMs) and linear dynamic
models (LDMs) as the acoustic model for automatic speech recognition
systems. In doing so, the individual strengths of both models, i.e.
the modelling of long-term temporal dependencies by the GMM-HMM