In-Vehicle Speech and Noise Corpora - Digital Signal Processing for In-Vehicle Systems and Safety - page 138

data collected during natural conversational interaction between the user and an
in-vehicle system. In the past, studies have analyzed the impact of in-vehicle noise
on speech systems including use of fixed noise and speech collection in lab
environments without the variability induced in either speech or noise. Recently,
some studies like [ 1 ] by Kawaguchi et al. have incorporated these variations.
Their corpus focuses on spontaneous conversational Japanese where the speech
data was collected under car idling and driving conditions. This study does
not include the environment variability of the speech due to the task-induced
stress. CU-Move focuses on compiling these variations in speech under diverse
acoustic conditions in the car environment along with various environments
encountered in realistic driving task. This data was collected from six different
vehicles. The core of this corpus includes over 300 speakers from six US cities, with
five speech style scenarios including route navigation dialogs. The noise collected
during this corpus identified over 14 different unique noise scenarios observed in
the car environment.
The goal of CU-Move is to enable the development of algorithms and technology
for robust access and transmission of information via spoken dialog systems in
mobile, hands free environments. The novel aspects of CU-Move include
corpora collection using microphone arrays on corpus development on speech
and acoustic vehicle conditions. This setup enables research utilizing environ-
mental classification for changing in-vehicle noise conditions and back-end
dialog navigation information retrieval subsystem connected to the WWW.
While previous attempts at in-vehicle speech systems have generally focused
on isolated command words to set radio frequencies, temperature control, etc.,
the CU-Move system is focused on natural conversational interaction between
the user and in-vehicle system. Since previous studies in speech recognition
have shown significant losses in performance when speakers are under task or
emotional stress, it is important to develop conversational systems that minimize
operator stress for the driver. System advances using CU-Move include intelligent
microphone arrays, auditory- and speaker-based constrained speech enhancement
methods, environmental noise characterization, and speech recognizer model adap-
tation methods for changing acoustic conditions in the car.
Here, the focus will be on the UTD-VN corpus with mention of relevant aspects in
the CU-Move corpus. In conjunction, these corpora address most of the variations
in the environment and speech encountered for holistic development of in-car speech
and communication systems.
9.2 The UT-Dallas Vehicle Noise Corpora
In the UTD-VN corpus, noise data samples were collected from 20 cars, five trucks,
and five SUVs across 10 different noise events. To enable portable recording across
vehicles, a portable, lightweight, high-fidelity data collection setup was used