Font Size: a A A

Study On Rapid Acoustic Event Detection In Noisy Driving Environment

Posted on:2015-07-28Degree:MasterType:Thesis
Country:ChinaCandidate:Q H ZhuFull Text:PDF
GTID:2298330422490876Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
Acoustic event is a single full continuous sound signal which can be easily perceived. Meanwhile, acoustic event detection means identifying the specific acoustic events by analyzing the observed audio features. Compared to traditional acoustic event detection, noisy driving environment has to deal with the following problems:driving environment belongs to open noisy environment,the main part in this noisy environment is the wind noise which is aerodynamic noise. The wind noise has a direct correlation with the vehicle, wind speed and the vehicle speed. But most of these factors can’t be predicted. Also, all of the acoustic event is relevant with human, so the detection of the system should be the same as the vote of human as far as possible, As well as the real-time is also a big challenge, real-time is an important factor that ensure road safety. If we can’t keep the read-time of the system, it will make a traffic accident.Faced with above challenges, we mainly study the rapid acoustic event detection methods in noisy driving environment, the main contents are as follows:Firstly, to effectively remove the noise in the observed signal and obtain a cleaner signal, we first model the noise in the signal and then we denoise the observed signal with spectral subtraction method. The concrete contents consist of these steps:analyze the correlation of the low frequency range and the high frequency range of the signal with the mutual information. model the noise with RBFNN where the input is the low frequency range and the output is the high frequency range.Secondly, to approach the identify effect of the human ear, we propose an acoustic event method based on the equal-loudness-level contour. As we know, human ear has different response for different frequency signal with the same energy. In order to eliminate the different and map the signal to real human auditory areas, we weight the observed signal with the A Weighting Filter which obtained from the equal-loudness-level contour. Then we can extract the features which are more compliance with human.Thirdly, to achieve the real-time of the algorithm, we put forward a method that get the trajectory characteristics of fundamental frequency based on that the trajectory of fundamental frequency of the specify acoustic event on the spectrogram is legible. The concrete steps are as follows:binary the spectrogram of the specify acoustic events; locate the frequency range of the observed signal with the binary spectrogram of the specify acoustic event; extract the main frequency of the observed signal. Compared to MFCC which base on the full range of the frequency, this method reduce amount of calculation, so it can raise the speed of the detection.The system is developed with C++programming language, and we also test for the system. The experiments approved that the system achieve the desired goals and have a good effect in robustness and real-time.
Keywords/Search Tags:Noisy Driving Environment, Acoustic event detection, RBFNN, equal-loudness-level contour, trajectory of fundamental frequency
PDF Full Text Request
Related items