Digital Signal Processing Reference
In-Depth Information
We evaluate the performance of our framework using an in-vehicle noise
database of 3 hours collected in 6 experimental runs using the same route and
the same vehicle on different days and hours. Fifteen noise classes are
transcribed during the data collection by a transcriber sitting in the car. The
time tags are generated instantly by the transcriber. After data collection,
some noise conditions are grouped together, resulting in 8 acoustically
distinguishable noise classes.
Figure 2-6.
Flow Diagram for In-Vehicle Environmental Sniffing
We identified the following primary noise conditions of interest: (N1- idle
noise consisting of the engine running with no movement and windows
closed, N2- city driving without traffic and windows closed, N3- city driving
with traffic and windows closed, N4- highway driving with windows closed,
N5-highway driving with windows 2 inches open, N6- highway driving with
windows half-way down, N7- windows 2 inches open in city traffic, NX-
others), which are considered as long term acoustic environmental conditions.
Other acoustic conditions (idle position with air-conditioning on, etc.) are
matched to these primary classes having the closest acoustic characteristic.
Since the Environmental Sniffing framework is not a speech system itself,
and must work with other speech systems, noise knowledge detection
performance for each noise type should be calculated by weighting each
