Learning Target Template for Acoustic Event Detection From Low-SNR Training Data

Abstract

Acoustic events of interest sometimes are inherently intermixed with strong background noise, like underwater sounds or mechanical fault signals. The noise usually creates a low-SNR scenario, making it more difficult to detect the event of interest. Suffering from lack of prior knowledge, it is a hard work to extract "correct" features from noisy acoustic mixtures for the target event. This paper addresses the detection problem for acoustic events based on low-SNR training data. A novel learning algorithm is proposed to extract the target's template from noisy training data. Based on a target-guided band-pass filter using wavelet packets, which has been proposed in a previous work for low-SNR event detection, the template of a specific event is represented in the context of the wavelet packet energy spectrum. Template learning is then formulated as an optimization problem maximizing the detection performance which is quantified as the output SNR of the filtered signal. It is further proved that the original complex optimization problem is equivalent to a linear program which is easy to solve. A relatively sufficient database was built to simulate the noise-polluted training signals and the low-SNR detection task. Results show that the learning method can work at -15 dB and above, and the trained template exhibit excellent detection performance comparable to the template extracted from clean samples. INDEX TERMSAcoustic event detection (AED), low SNR, noisy training data, template learning.

Sound Target Detection Under Noisy Environment Using Brain-Computer Interface

“…We saw that the detection rates of all eight subjects were high (all over 97%, with an average of 99.08%; the detection rates of Subjects 2, 4, 5, and 7 were 100%). The average RT of 4 subjects (1, 2, 4, 7) was around 0.7 seconds, whereas the RT of the other four subjects was about 0.5 seconds (3,5,6,8). For the successful trials, the standard deviation of RT was around 20-196 ms.…”

“…Since most STD studies focus on the condition of SNR (in most of the research on sound event detection, 0dB~-5dB is considered as the low level of SNR. For example, [4][5][6]), we calculated the SNR of the target sound relative to the background sound. As the background noise used in this study was the wind recorded in the field, the intensity of sound changed over time.…”

“…The performance of the model in the low SNR test set was much lower than in the high SNR test set (precision decreased by 36.51%, and recall decreased by 57.58%). Ren et al [5] demonstrated the influence of SNR on the detection performance in the study of the dangerous SED. In the same noise scene, the detection error rate of the same sound target on the test set reached 37.38% when the SNR was -15 dB and increased by 30.2% compared with the condition of -5 dB.…”